<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Tharushi  Karunaratne on Medium]]></title>
        <description><![CDATA[Stories by Tharushi  Karunaratne on Medium]]></description>
        <link>https://medium.com/@tharushi.cnnp?source=rss-de5983df3c8d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*K7OgrdNxO2WIFB9M7ag4aA.png</url>
            <title>Stories by Tharushi  Karunaratne on Medium</title>
            <link>https://medium.com/@tharushi.cnnp?source=rss-de5983df3c8d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 23:39:36 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@tharushi.cnnp/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Unlocking the Brain: How Ultrasound and Tiny Bubbles Are Rewriting the Rules of Brain Cancer…]]></title>
            <link>https://medium.com/@tharushi.cnnp/unlocking-the-brain-how-ultrasound-and-tiny-bubbles-are-rewriting-the-rules-of-brain-cancer-6d56aaa646dc?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/6d56aaa646dc</guid>
            <category><![CDATA[focused-ultrasound]]></category>
            <category><![CDATA[medical-innovation]]></category>
            <category><![CDATA[neurotechnology]]></category>
            <category><![CDATA[biotechnology]]></category>
            <category><![CDATA[neuroscience]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Wed, 11 Feb 2026 16:30:29 GMT</pubDate>
            <atom:updated>2026-02-11T16:30:29.679Z</atom:updated>
            <content:encoded><![CDATA[<h3>Unlocking the Brain: How Ultrasound and Tiny Bubbles Are Rewriting the Rules of Brain Cancer Treatment</h3><p><em>The challenge of treating glioblastoma — and how low intensity focused ultrasound is changing the game</em></p><p>Glioblastoma multiforme represents one of medicine’s most frustrating challenges. These aggressive brain tumors are notoriously difficult to treat, not because we lack effective drugs, but because we can’t get those drugs where they need to go.</p><p>The problem? The blood-brain barrier — our brain’s natural defense system that keeps harmful substances out. Unfortunately, it also keeps out the chemotherapy drugs that could save lives.</p><h3>The Double-Barrier Problem</h3><p>Think of glioblastoma treatment as facing two locked doors:</p><p><strong>The Blood-Brain Barrier (BBB):</strong> This is your brain’s security system. It consists of tightly packed cells forming the walls of blood vessels in the brain. These cells are held together by “tight junctions” — molecular locks that prevent almost anything in your bloodstream from entering your brain tissue. Additionally, specialized proteins called P-glycoproteins act like bouncers, actively ejecting any substances that manage to slip through.</p><p><strong>The Blood-Tumor Barrier (BTB):</strong> As glioblastomas grow rapidly, they create their own chaotic network of blood vessels. While these vessels are somewhat “leaky,” they’re still protected by the same mechanisms as healthy brain tissue. Worse, the tumor’s rapid growth creates high internal pressure that pushes drugs back out, even when they manage to get in.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/675/1*cP9KImFwPcy_rv95ncw6qw.jpeg" /><figcaption>Blood Brain Barrier and Blood Tumor Barrier — <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC9169875/">https://pmc.ncbi.nlm.nih.gov/articles/PMC9169875/</a></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AfZoYJSQVdB4790aGt0A_g.png" /></figure><p>The result? Even after surgery removes the visible tumor core, cancer cells infiltrate about 2 centimeters into surrounding healthy tissue — an area impossible to remove without causing severe brain damage. Without reaching these invasive cells with chemotherapy, recurrence is almost inevitable.</p><h3>Enter LIFU: A Non-Invasive Solution</h3><p>This is where Low-Intensity Focused Ultrasound (LIFU) becomes game-changing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hHcTNprpDVInCUXZcPWUhw.png" /></figure><p>Unlike high-intensity ultrasound used to destroy tissue through heat, LIFU uses sound waves at three orders of magnitude lower power — gentle enough not to damage tissue, but precise enough to temporarily open the blood-brain barrier exactly where needed.</p><h4>How It Works: The Microbubble Dance</h4><p>The magic happens when LIFU teams up with tiny gas-filled bubbles called microbubbles:</p><ol><li><strong>Injection:</strong> Microbubbles (about 1–10 micrometers in diameter) are injected into the bloodstream. These have a perfluorocarbon gas core wrapped in a lipid shell — think of them as molecular-sized bubble wrap that circulates safely for over 30 minutes.</li><li><strong>Targeting:</strong> Focused ultrasound waves converge on the exact brain region where treatment is needed — with millimeter precision.</li><li><strong>Activation:</strong> When ultrasound waves hit the microbubbles, something fascinating happens. The bubbles oscillate — expanding during the low-pressure phase of the sound wave and contracting during the high-pressure phase. This creates mechanical forces that temporarily reorganize the barrier.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F9HCCLFsBtGEkQ2QTRqPdA.png" /></figure><h4>The Molecular Choreography</h4><p>This isn’t crude demolition — it’s precise molecular manipulation:</p><ul><li><strong>Stable Cavitation</strong> (at lower pressures): Microbubbles oscillate smoothly and symmetrically, creating fluid currents that exert controlled shear stress on blood vessel walls</li><li><strong>Tight Junction Modulation:</strong> The mechanical forces temporarily redistribute proteins like claudin-5, occludin, and ZO-1 — essentially unlocking the molecular gates</li><li><strong>P-Glycoprotein Suppression:</strong> Those molecular bouncers? Their expression gets dialed down for up to 48 hours, extending the therapeutic window</li><li><strong>Enhanced Transport:</strong> Both paracellular (between cells) and transcellular (through cells) pathways open up</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tr3mFasDyYxXF0K787h5sQ.png" /></figure><p>Most remarkably, this is <strong>reversible</strong>. The barrier begins restoring integrity almost immediately, with complete closure typically occurring within 6–24 hours.</p><h4>Engineering the Perfect Treatment</h4><p>Getting this right requires fine-tuning multiple parameters:</p><p><strong>Frequency:</strong> Typically around 1 MHz — low enough to penetrate skull bone effectively <strong>Acoustic Pressure:</strong> The sweet spot is 0.31–0.84 MPa. Too low, and nothing happens. Too high, and you risk tissue damage <strong>Duty Cycle:</strong> Usually pulsed (0.5% or 3.33%) to allow microbubbles time to circulate and replenish between bursts <strong>Mechanical Index (MI):</strong> A critical safety metric that must be carefully monitored</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*i9tpvFbOdm9cVpdL4UYyig.png" /></figure><p>The challenge of penetrating the skull is solved using <strong>Magnetic Resonance-guided Focused Ultrasound (MRgFUS)</strong> — real-time MRI navigation that provides millimeter-precise targeting and temperature monitoring.</p><h3>Seeing Results: Verification and Diagnostics</h3><p>After treatment, doctors verify success using contrast-enhanced MRI. Because gadolinium contrast agents normally can’t cross the blood-brain barrier, seeing them in brain tissue on post-procedure scans confirms the barrier is open — like checking that a door is unlocked.</p><h3>The Two-Way Street: Liquid Biopsy</h3><p>Here’s an unexpected bonus: when the barrier opens, traffic flows both ways. Tumor biomarkers that were trapped in the brain can now escape into the bloodstream, enabling:</p><ul><li>Detection of cell-free DNA (cfDNA)</li><li>Brain-specific proteins like S100b</li><li>Mutant copies of isocitrate dehydrogenase 1 (IDH-1)</li></ul><p>This creates a non-invasive “liquid biopsy” option for molecular profiling — no brain surgery required.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qzf6uBRPrWDxWZTQyKISgQ.png" /></figure><h3>Clinical Reality: From Lab to Patient</h3><p>After more than 20 years in preclinical development, LIFU+microbubbles is now in clinical trials. The FDA has already approved the technology for essential tremor and Parkinson’s disease, and glioblastoma applications are advancing rapidly.</p><p>Early clinical results are encouraging:</p><ul><li><strong>Safety:</strong> Well-tolerated across multiple sessions with no neurotoxicity</li><li><strong>Reversibility:</strong> Barrier disruption resolves within 20–24 hours</li><li><strong>Efficacy:</strong> Patients with successful barrier opening showed increased progression-free survival and median overall survival</li><li><strong>Precision:</strong> 92.4% accuracy in targeting when using both T1 and T2 weighted MRI imaging</li></ul><h3>Therapeutic Delivery: Proof of Concept</h3><p>Preclinical studies have demonstrated remarkable improvements when LIFU enhances drug delivery:</p><ul><li><strong>Etoposide:</strong> 8-fold increase in brain tumor concentration</li><li><strong>Doxorubicin:</strong> 4–50x increase (depending on delivery method)</li><li><strong>Carboplatin:</strong> 2.9–6x increase with corresponding survival improvements</li><li><strong>Paclitaxel:</strong> 3–5x increase with significant tumor control</li></ul><p>These aren’t marginal improvements — they’re transformative changes in how much medication reaches the tumor.</p><h3>Safety Considerations</h3><p>At optimal parameters, LIFU+microbubbles shows an excellent safety profile:</p><p>✓ No changes in vital signs or motor coordination</p><p>✓ Normal blood cell counts maintained</p><p>✓ No liver, kidney, or cardiac dysfunction</p><p>✓ Histology shows no tissue damage when parameters are controlled</p><p>✓ Minimal inflammatory response that resolves quickly</p><p>The key is staying within the therapeutic window. Exceeding 0.85 MPa acoustic pressure can cause red blood cell extravasation (petechiae), and going much higher risks hemorrhage.</p><h3>The Future Is Sound</h3><p>We’re witnessing a paradigm shift in how we might treat not just glioblastoma, but potentially many central nervous system diseases. LIFU+microbubbles offers what seemed impossible: precise, temporary, repeatable, and safe blood-brain barrier opening.</p><p>For patients facing glioblastoma — a diagnosis that currently carries a median survival of 15–20 months — this technology represents hope. It transforms drugs that were previously ineffective (not because they didn’t work, but because they couldn’t reach their target) into potential treatments.</p><p>The implications extend beyond glioblastoma: Alzheimer’s disease, Parkinson’s disease, other brain tumors, neurological infections — any condition where the blood-brain barrier prevents effective treatment could potentially benefit.</p><h3>What Makes This Revolutionary</h3><p>Let’s be clear about what makes LIFU+microbubbles special:</p><ul><li><strong>Non-invasive:</strong> No surgery, no implants</li><li><strong>Precise:</strong> Millimeter-accurate targeting via MRI guidance</li><li><strong>Reversible:</strong> Barrier restores itself naturally within hours</li><li><strong>Repeatable:</strong> Multiple treatment sessions are safe</li><li><strong>Controllable:</strong> Real-time monitoring and adjustment possible</li><li><strong>Versatile:</strong> Can deliver drugs of various sizes and types</li></ul><p>Traditional approaches — intrathecal injection, convection-enhanced delivery, surgical implants — all involve invasive procedures with risks of infection, hemorrhage, and inconsistent drug distribution. LIFU+microbubbles potentially makes all of that unnecessary.</p><h3>The Bottom Line</h3><p>The blood-brain barrier exists to protect us. For over a century since its discovery, it’s been both our defender and our obstacle. Low-intensity focused ultrasound with microbubbles doesn’t destroy this barrier — it negotiates with it, opening a temporary doorway precisely when and where we need it.</p><p>For the hundreds of thousands of people diagnosed with glioblastoma and other brain diseases each year, this could be the breakthrough that transforms their prognosis. The door that was locked is opening — and hope is walking through.</p><p>Source —</p><ul><li><a href="https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2022.903059/full">https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2022.903059/full</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6d56aaa646dc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Seeing Inside the Brain: How Light Could Save Stroke Patients]]></title>
            <link>https://medium.com/@tharushi.cnnp/seeing-inside-the-brain-how-light-could-save-stroke-patients-670bc7aee800?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/670bc7aee800</guid>
            <category><![CDATA[neurotechnology]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[clinical-engineering]]></category>
            <category><![CDATA[signal-processing]]></category>
            <category><![CDATA[healthtech]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Wed, 04 Feb 2026 02:09:38 GMT</pubDate>
            <atom:updated>2026-02-04T02:09:38.675Z</atom:updated>
            <content:encoded><![CDATA[<p>Every second counts when someone has a stroke. Neurologists have a saying: “Time is brain” — because with each passing second that blood flow to the brain is compromised, millions of neurons die irreversibly. Stroke and traumatic brain injury massively disrupt cerebral blood supply, causing permanent damage and devastating outcomes.</p><p>One major reason for this permanent damage? The gap between when the incident occurs and when doctors can reliably diagnose it. Currently, strokes are diagnosed clinically and later confirmed through imaging techniques like CT or MRI. The gold standard for measuring cerebral blood flow — O-15 positron emission tomography (PET) — is expensive, exposes patients to ionizing radiation, and only provides a snapshot rather than continuous monitoring.</p><p>But what if we could monitor cerebral blood flow continuously, in real-time?</p><p>That’s exactly what the <strong>Openwater wearable optical system</strong> aims to deliver.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Vqw2XJNhdjepI1kaccpeJg.png" /></figure><h3>The Openwater Architecture: A Wearable Brain Monitor</h3><p>The Openwater system is elegantly designed with two near-infrared light sources that measure two critical parameters of cerebral circulation:</p><ul><li><strong>Relative blood flow (rBF)</strong></li><li><strong>Relative blood volume (rBV)</strong></li></ul><p>The device was tested during breath-holding maneuvres — a simple technique that physiologically increases cerebral blood flow by raising carbon dioxide levels (hypercapnia).</p><p>The magic happens through something called <strong>laser speckle contrast</strong>. A specialized CMOS camera, positioned 36 mm away from each light source, continuously captures these speckle patterns. From these patterns:</p><ul><li><strong>Relative blood flow</strong> is derived from fluctuations in the laser speckle</li><li><strong>Relative blood volume</strong> is derived from changes in the speckle’s average intensity</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2mx1nU_m6aQEzh9wH8Pkiw.png" /><figcaption>Experimental setup and raw time-series data — <a href="https://pubmed.ncbi.nlm.nih.gov/38464864/">https://pubmed.ncbi.nlm.nih.gov/38464864/</a></figcaption></figure><h3>The Technology Behind the Magic</h3><h4>What Exactly Is a Laser Speckle?</h4><p>When laser light enters the brain, it undergoes multiple scattering events within brain tissue, creating a random interference pattern called a “laser speckle.” The device’s custom camera, positioned 36 mm away, continuously captures these speckle intensity fluctuations — which are directly related to the movement of red blood cells.</p><p>Think of it like this: as blood cells move through your brain’s vessels, they create tiny disturbances in the laser light. These disturbances form patterns that the camera can detect and interpret.</p><h4>Measuring Blood Flow and Volume</h4><p><strong>Relative Blood Flow (rBF):</strong> This metric provides a relative index of microvascular blood flow, derived from fluctuations in speckle contrast — a property of the laser light pattern created by moving red blood cells.</p><p><strong>Relative Blood Volume (rBV):</strong> This metric tracks changes in the amount of blood in the tissue, derived from changes in the speckle average intensity, or the overall brightness of the detected light.</p><h3>The Engineering Challenge: Getting Light Deep Into the Brain</h3><h4>Breaking Through the Skull Barrier</h4><p>Historically, the major bottleneck in transcranial optical monitoring has been the skull’s high scattering coefficient, which typically traps photons in superficial layers. The Openwater system cleverly overcomes this with a specific source-detector separation geometry designed to bias photon paths toward the cerebral cortex.</p><p>While many Diffuse Correlation Spectroscopy (DCS) studies use a standard 25 mm separation, the Openwater system extends this to <strong>36 mm</strong>. This wider geometry creates a “banana-shaped” diffusive transport path that reaches deeper into brain tissue. By increasing the separation, the system ensures that detected photons have sampled the microvasculature of the frontal lobe rather than just the scalp.</p><h4>The 785 nm Sweet Spot</h4><p>The choice of 785 nm wavelength is mathematically significant — it represents the <strong>isosbestic point</strong> for hemoglobin. At this wavelength, oxygenated and deoxygenated hemoglobin absorb light almost identically. This means changes in the optical signal are “pure” measures of hemodynamics — specifically red blood cell movement and volume — unaffected by fluctuations in arterial oxygen saturation.</p><p>However, this 36 mm distance creates a significant signal-to-noise challenge, as the number of returning photons is drastically reduced. This is where the system’s unique temporal gating and parallelized detection strategies come in.</p><h3>Maximizing Sensitivity: The Power of Parallelization</h3><p>At 36 mm source-detector separations, the primary obstacle is “read noise” from the CMOS sensor. The Openwater system solves this by moving away from the single-photon detection used in traditional DCS and toward <strong>parallelized homodyne detection</strong>.</p><p>Here’s how it works:</p><p>The system uses a <strong>5-megapixel CMOS sensor</strong> to sample 5 million coherence areas (speckles) simultaneously. This “Speckle Ensemble Detection” allows statistical averaging of millions of pixels for every 250 µs pulse. To isolate the physiological signal, the system applies variance correction and subtracts a contrast offset to account for pixel non-uniformity.</p><p>The result? A remarkably stable signal with only <strong>0.2% variation</strong> in static phantoms.</p><p>This massive parallelization allows the device to function at separations where traditional DCS would be overwhelmed by low photon counts. By sampling millions of speckles at once, the Openwater system achieves the signal-to-noise ratio necessary to resolve individual heartbeats in real-time — even through the dense adult skull.</p><h3>The 250 Microsecond Advantage</h3><h4>Why Pulse Length Matters</h4><p>In optical flowmetry, temporal gating is critical for maximizing sensitivity to deep tissue flow. Traditional continuous-wave systems can’t easily distinguish between slow decorrelation of surface tissue and the rapid dynamics of the brain.</p><p>The Openwater system uses <strong>250 µs pulse width</strong>. First-principles analysis reveals that the sensitivity peak — defined as the maximum derivative of contrast with respect to flow — occurs at approximately 200 µs for adult cerebral geometries. The system uses 250 µs to balance this sensitivity peak with the need for sufficient photon counts.</p><p>Why are shorter pulses necessary? Because photons in the diffusive transport regime (multiply scattered light) decorrelate significantly faster than single-scattered light from the surface.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KMIMVfNr8V009eSObnBGEQ.png" /></figure><h4>The Technical Implementation</h4><p>This is executed via a <strong>Master Oscillator Power Amplifier (MOPA)</strong> configuration. The master oscillator provides a highly coherent seed signal, which is amplified to a high peak power of <strong>1.6 W per channel</strong>. With a 1% duty cycle, the system maintains a safe average power of 16 mW while delivering high energy (400 µJ) per pulse.</p><p><strong>The Bottom Line:</strong> Standard Laser Speckle Contrast Imaging typically uses exposure times of &gt;1 ms, optimized for surface flow. By shifting to the 250 µs regime, the Openwater system achieves a <strong>&gt;3x increase in waveform amplitude</strong> compared to a 1000 µs pulse. This temporal precision allows the device to resolve fast-moving dynamics of deep cerebral vasculature that continuous-wave systems would simply average out as noise.</p><h3>Capturing the Heartbeat of the Brain</h3><p>The clinical utility of the system is demonstrated by its ability to resolve the <strong>pulsatile flow waveform</strong>, which characterizes cerebrovascular resistance and stiffness. The raw 40 Hz data is interpolated to 125 Hz to enable synchronization with gold-standard Transcranial Doppler (TCD).</p><p>The system successfully captures three distinct hemodynamic peaks:</p><ol><li><strong>P1 (Percussion Wave):</strong> Reflects left ventricular ejection</li><li><strong>P2 (Tidal Wave):</strong> The pulse wave reflected by the closing aortic valve</li><li><strong>P3 (Dicrotic Wave):</strong> Represents diastolic flow</li></ol><p>The ability to resolve the dicrotic notch and individual peaks provides “face validity” for the device’s accuracy. By mirroring the morphological features of TCD — the clinical standard — the Openwater system proves it can accurately “see” the cerebral vasculature using light instead of sound.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qxZfwTH9W-gR6VU_xOE6ag.png" /></figure><h3>Validation: Putting the Technology to the Test</h3><p>In a study of 25 healthy adults, researchers compared the Openwater system against transcranial Doppler ultrasound during breath-holding maneuvers (which increase cerebral blood flow through hypercapnia).</p><h4>The Results Were Impressive:</h4><ul><li><strong>Strong correlation</strong> between optical blood flow and TCD measurements: R = 0.79</li><li><strong>Beat-to-beat correlation</strong> for individual subjects averaged 0.88</li><li>The linear fit slope of 0.87 (95% CI: 0.83–0.92) indicated excellent agreement</li><li><strong>Blood volume changes</strong> were also correlated (R = 0.72), though appropriately smaller than flow changes — exactly as expected from cardiovascular physiology</li></ul><p>The study also showed strong agreement in:</p><ul><li>Breath-hold index calculations</li><li>Timing of maximum hemodynamic effects</li><li>Waveform morphology features (P1, P2, P3, and dicrotic notch)</li><li>Pulsatility index changes</li><li>Augmentation index (vascular stiffness measure)</li></ul><h3>Real-World Applications: The Future of Stroke Care</h3><p>This technology could revolutionize neurological care in several ways:</p><h4>1. Bedside Monitoring for Acute Ischemic Stroke</h4><p>Continuous, real-time monitoring could help optimize perfusion in stroke patients, guiding treatment decisions minute by minute.</p><h4>2. Non-Invasive Detection of Large Vessel Occlusions</h4><p>Early detection of blocked arteries could dramatically reduce time to treatment.</p><h4>3. Continuous Assessment of Intracranial Pressure</h4><p>The system could monitor elevated intracranial pressure and impaired compliance without invasive sensors.</p><h4>4. Personalized Stroke Treatment</h4><p>Real-time hemodynamic data could enable truly personalized, physiology-guided care rather than one-size-fits-all protocols.</p><h3>Why This Matters</h3><p>Unlike current monitoring techniques, the Openwater system offers several advantages:</p><ul><li><strong>Non-invasive:</strong> No radiation exposure, no surgical implantation</li><li><strong>Continuous:</strong> Real-time monitoring instead of single snapshots</li><li><strong>Portable:</strong> Wearable headset instead of room-sized machines</li><li><strong>User-friendly:</strong> No need for specialized ultrasound technicians</li><li><strong>Accessible:</strong> Works on patients who lack adequate ultrasound windows (nearly 20% of the population)</li></ul><p>The device is particularly valuable because it measures actual tissue-level blood flow in the brain’s microvasculature, not just velocity in large arteries.</p><h3>A New Window Into the Brain</h3><p>The Openwater wearable optical system represents a significant step forward in non-invasive brain monitoring. By harnessing the physics of laser speckle contrast and sophisticated engineering to overcome the skull barrier, this technology offers a practical way to continuously monitor cerebral blood flow at the bedside.</p><p><em>What do you think about this technology? Could wearable brain monitors become as common as heart rate monitors? Share your thoughts in the comments below.</em></p><p><a href="https://pubmed.ncbi.nlm.nih.gov/38464864/">Validation of the Openwater wearable optical system: cerebral hemodynamic monitoring during a breath-hold maneuver - PubMed</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=670bc7aee800" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Perceptron: How a 1950s Algorithm Sparked the AI Revolution]]></title>
            <link>https://medium.com/@tharushi.cnnp/the-perceptron-how-a-1950s-algorithm-sparked-the-ai-revolution-b5285c5eb808?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/b5285c5eb808</guid>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[perceptron]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[neurotechnology]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Thu, 01 Jan 2026 18:23:27 GMT</pubDate>
            <atom:updated>2026-01-01T18:23:27.106Z</atom:updated>
            <content:encoded><![CDATA[<p>The human brain — three pounds of biological tissue containing roughly 86 billion neurons — has fascinated scientists for centuries. How does this electrochemical network give rise to thought, memory, and consciousness? Rather than merely studying the brain from the outside, pioneering scientists in the 1940s and 50s took a bold approach: <em>build artificial versions and see if they work</em>.</p><p>This audacious idea didn’t just advance neuroscience. It laid the foundation for modern artificial intelligence.</p><h3>The First Artificial Neurons</h3><p>In 1943, neurophysiologist Warren McCulloch and logician Walter Pitts published a landmark paper proposing a mathematical model of how neurons might compute. Their “McCulloch-Pitts neuron” was elegantly simple: it took binary inputs (0s and 1s), performed a logical operation, and produced a binary output.</p><p>Think of it like a light switch that turns on only when enough other switches are flipped. If the sum of inputs exceeded a threshold, the neuron “fired” (output 1); otherwise, it stayed silent (output 0).</p><p>This model inspired <a href="https://medium.com/@tharushi.cnnp/snarc-the-1951-machine-that-taught-itself-to-navigate-mazes-a5ddb6daddcc">SNARC </a>(Stochastic Neural Analog Reinforcement Calculator), built by Marvin Minsky and Dean Edmonds in 1951 — one of the first neural network machines ever constructed. But the McCulloch-Pitts neuron had a critical limitation: <strong>it couldn’t learn</strong>. Its thresholds were fixed. It couldn’t adapt from experience.</p><h3>Enter the Perceptron</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WCMtNra6_dSaLpnSpkEexA.png" /></figure><p>In 1957, Cornell psychologist Frank Rosenblatt changed everything. He introduced the <strong>Perceptron</strong>, an artificial neuron that could actually learn from examples. Funded by the U.S. Office of Naval Research, Rosenblatt built a hardware implementation called the Mark I Perceptron, designed to recognize simple visual patterns.</p><p>The perceptron was genuinely revolutionary. Here’s how it works.</p><h3>Anatomy of a Perceptron</h3><p>Imagine you’re deciding whether to go to the beach. You consider three factors:</p><ul><li><strong>Weather</strong> (sunny or rainy)</li><li><strong>Temperature</strong> (hot or cold)</li><li><strong>Work schedule</strong> (busy or free)</li></ul><p>Each factor has a different importance to you. Maybe weather matters most, temperature less so, and your schedule is somewhere in between. A perceptron makes decisions exactly this way.</p><h3>The Architecture</h3><p>A perceptron consists of:</p><ol><li><strong>Input layer</strong>: Features or attributes (x₁, x₂, x₃, …)</li><li><strong>Weights</strong>: Importance values for each input (w₁, w₂, w₃, …)</li><li><strong>Bias</strong>: A threshold that shifts the decision boundary</li><li><strong>Activation function</strong>: Converts the weighted sum into an output</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cGr-hhAatFzHJ8yed7uAnA.png" /><figcaption>Architecture of the Perceptron — image credits — <br><strong>Perceptron | Neural Networks by </strong><a href="https://www.youtube.com/@firstprinciplesofcomputerv3258">First Principles of Computer Vision</a></figcaption></figure><h3>The Mathematics</h3><p>Here’s the step-by-step process:</p><p><strong>Step 1: Weighted Sum</strong></p><p>The perceptron calculates a weighted sum of inputs, then subtracts the bias:</p><pre>z = (x₁ × w₁) + (x₂ × w₂) + (x₃ × w₃) + ... - b</pre><p>Or more compactly: z = Σ(xᵢ × wᵢ) - b</p><p><strong>Step 2: Activation Function</strong></p><p>The result passes through a step function (also called the Heaviside function):</p><pre>f(z) = {<br>  1, if z &gt; 0<br>  0, if z ≤ 0<br>}</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-wQHH4jdn8AyaysAwTv9vg.png" /><figcaption>Activation function — image credits — <br><strong>Perceptron | Neural Networks by </strong><a href="https://www.youtube.com/@firstprinciplesofcomputerv3258">First Principles of Computer Vision</a></figcaption></figure><p>This binary output mirrors the “all-or-none principle” in biological neurons — they either fire an action potential or they don’t. There’s no middle ground.</p><h3>A Concrete Example</h3><p>Let’s classify whether an email is spam:</p><ul><li><strong>Inputs</strong>: x₁ = contains “free” (1 if yes, 0 if no), x₂ = contains “urgent” (1/0), x₃ = from known contact (1/0)</li><li><strong>Weights</strong>: w₁ = 0.7, w₂ = 0.6, w₃ = -0.9 (negative because known contacts reduce spam probability)</li><li><strong>Bias</strong>: b = 0.5</li></ul><p>For an email with “free” and “urgent” from an unknown sender:</p><ul><li>x₁ = 1, x₂ = 1, x₃ = 0</li></ul><p>Calculate z:</p><pre>z = (1 × 0.7) + (1 × 0.6) + (0 × -0.9) - 0.5<br>z = 0.7 + 0.6 + 0 - 0.5 = 0.8</pre><p>Since z &gt; 0, the output is <strong>1</strong> (spam).</p><h3>The Learning Algorithm</h3><p>What makes perceptrons special isn’t just making predictions — it’s <em>learning</em> to make better predictions.</p><p>The perceptron learning algorithm:</p><ol><li>Initialize weights randomly</li><li>For each training example:</li></ol><ul><li>Make a prediction</li><li>Calculate the error (true label — predicted label)</li><li>Update weights: wᵢ = wᵢ + learning_rate × error × xᵢ</li></ul><p>3. Repeat until errors are minimized.</p><p>This is <strong>supervised learning</strong> in its purest form. Show the perceptron examples with correct answers, and it adjusts its weights to reduce mistakes.</p><h3>The Geometric Intuition</h3><p>Here’s the beautiful insight: a perceptron is a <strong>linear classifier</strong>. In two dimensions, it draws a straight line separating two classes. In three dimensions, a plane. In higher dimensions, a hyperplane.</p><p>The equation w₁x₁ + w₂x₂ - b = 0 defines this boundary. Points on one side are classified as 1, points on the other as 0. Learning means finding the right orientation and position for this boundary.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RhIBKmArSR-KlC0Wk7MJKw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ldNpK6Jl_z_QaXEP8grzjA.png" /></figure><p>This foundation led directly to linear regression, logistic regression, and support vector machines — workhorses of modern machine learning.</p><h3>The AI Winter and Perceptron Limitations</h3><p>In 1969, Marvin Minsky and Seymour Papert published “Perceptrons,” a book that proved a devastating mathematical limitation: <strong>perceptrons cannot solve non-linearly separable problems</strong>.</p><p>The classic example is XOR (exclusive OR):</p><pre>Input 1 | Input 2 | Output<br>   0    |    0    |   0<br>   0    |    1    |   1<br>   1    |    0    |   1<br>   1    |    1    |   0</pre><p>No single straight line can separate the 1s from the 0s in this pattern. The perceptron fails completely.</p><p>This revelation, combined with hardware limitations and funding disappointments, triggered the first “AI Winter” — a period of reduced research and pessimism lasting over a decade. The field nearly died.</p><h3>The Phoenix Rises: Modern Neural Networks</h3><p>But the story didn’t end there. The solution? <strong>Stack perceptrons in layers</strong>.</p><p>Multi-layer perceptrons (neural networks) can learn non-linear boundaries by combining multiple linear classifiers. Add non-linear activation functions (sigmoid, ReLU), and suddenly you can approximate any function — a property called universal approximation.</p><p>Every modern neural network, from image classifiers to large language models like GPT, traces its lineage back to Rosenblatt’s perceptron. The humble single-layer classifier that couldn’t solve XOR evolved into architectures with billions of parameters solving problems once thought impossible.</p><h3>Why This Matters Today</h3><p>The perceptron teaches us three profound lessons:</p><p><strong>1. Simple mechanisms, complex behavior</strong>: Individual neurons are simple threshold devices, yet networks of them produce intelligence.</p><p><strong>2. Learning from data</strong>: The perceptron pioneered the idea that machines don’t need explicit programming — they can learn patterns from examples.</p><p><strong>3. Limitations drive innovation</strong>: Understanding what perceptrons <em>couldn’t</em> do led to the breakthroughs that define modern AI.</p><h3>Conclusion</h3><p>When Frank Rosenblatt demonstrated the Mark I Perceptron in 1960, identifying simple shapes on a 20×20 photocell array, the technology press went wild. The hype eventually led to disappointment, then to the AI winter.</p><p>But Rosenblatt’s core insight was correct: artificial neurons that learn from experience could be the path to machine intelligence. He just couldn’t have imagined that his algorithm would need to wait decades for sufficient computing power, massive datasets, and architectural innovations to fulfill its promise.</p><p>Today, as you use facial recognition to unlock your phone or ask ChatGPT a question, you’re witnessing the culmination of that 1957 breakthrough. The perceptron didn’t just spark the AI revolution — it <em>was</em> the spark. Everything else is just scaling up the fire.</p><p><a href="https://www.researchgate.net/publication/378940739_Introduction_to_the_Perceptron_and_Its_Applications">https://www.researchgate.net/publication/378940739_Introduction_to_the_Perceptron_and_Its_Applications</a></p><p><a href="https://pubmed.ncbi.nlm.nih.gov/31550405/">https://pubmed.ncbi.nlm.nih.gov/31550405/</a></p><p><a href="https://www.youtube.com/watch?v=OFbnpY_k7js&amp;t=266s">https://www.youtube.com/watch?v=OFbnpY_k7js&amp;t=266s</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b5285c5eb808" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[SNARC: The 1951 Machine That Taught Itself to Navigate Mazes]]></title>
            <link>https://medium.com/@tharushi.cnnp/snarc-the-1951-machine-that-taught-itself-to-navigate-mazes-a5ddb6daddcc?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/a5ddb6daddcc</guid>
            <category><![CDATA[aihistory]]></category>
            <category><![CDATA[neuromorphic-computing]]></category>
            <category><![CDATA[reinforcement-learning]]></category>
            <category><![CDATA[innovation]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Fri, 26 Dec 2025 06:28:50 GMT</pubDate>
            <atom:updated>2025-12-26T06:28:50.937Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hS9IwxOmNyw2AaJjRFn6rQ.png" /><figcaption>Image credits — Notebooklm</figcaption></figure><p>In 1951, while most computers were still glorified calculators executing rigid sequences of instructions, two graduate students at Princeton built something radically different: a machine that could learn from experience.</p><p>The Stochastic Neural Analog Reinforcement Computer — mercifully abbreviated to SNARC — wasn’t just another early computer. It was the world’s first hardware implementation of a neural network, and arguably the first neuromorphic system ever constructed. Built by Marvin Minsky and Dean Edmonds, SNARC represented a bold attempt to understand intelligence not by programming it, but by recreating the physical processes that give rise to it.</p><h3><strong>The Question That Started It All</strong></h3><p>Minsky was driven by what he considered the most profound mystery in science: the human mind. Rather than treating the brain as an impenetrable black box, he asked a deceptively simple question: Could we build a machine that learns the way living creatures do?</p><p>The inspiration came from a landmark 1943 paper by Warren McCulloch and Walter Pitts, which proposed that neural activity could be understood through mathematical logic. But Minsky took this idea further. He didn’t want to simulate neurons on paper — he wanted to build them in hardware, creating a physical system that could adapt and improve through trial and error.</p><p>This was the essence of connectionism: the idea that intelligence emerges not from symbolic manipulation or explicit programming, but from the interactions of simple, interconnected units that modify their behavior based on experience.</p><h3>Building a Brain from Bomber Parts</h3><p>The SNARC was an analog machine built around 40 artificial neurons, each implementing what we now call Hebbian synapses. Every synapse maintained a “memory” — a probability value between 0 and 1 — that determined the likelihood of an incoming signal triggering an outgoing pulse. This stochastic architecture reflected something fundamental about biological learning: it’s inherently probabilistic, not deterministic.</p><p>But here’s where things get interesting. To implement the learning mechanism, Minsky needed a way to physically adjust those probability values in response to success or failure. His solution was ingeniously makeshift: a motor salvaged from a Honeywell C-1 gyroscopic autopilot, the kind used to keep B-24 bombers flying straight during World War II. This motor drove a chain connected to all 40 synapses, each equipped with a potentiometer that could adjust its firing probability.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_z-oGHxcPlt613fH_kbQ1w.png" /><figcaption>Image credits — Notebooklm</figcaption></figure><p>The learning process itself was pure behavioral psychology made tangible. Minsky had consulted extensively with B.F. Skinner, the father of operant conditioning, to ensure the hardware faithfully embodied principles of reinforcement learning. The machine was tested on a copy of Claude Shannon’s maze — a fitting tribute to another pioneer who saw computation as more than mere number-crunching.</p><h3><strong>How a Machine Learns to Escape a Maze</strong></h3><p>The learning mechanism was elegantly simple. When a neuron fired, a capacitor would hold a charge for several seconds, creating a short-term memory trace of recent activity. As the simulated rat wandered through the maze, this trace would fade naturally, like a scent trail growing cold.</p><p>When the rat reached the goal, the human operator pressed a “reward” button. This triggered the motor, which began turning the chain drive. But here’s the clever part: each synapse had an electromagnetic clutch that would only engage if its capacitor was currently charged. This meant only the neurons that had fired recently — those involved in the successful path — would have their probabilities increased.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RBvLmXD3yPLFGvfrUJ7jZg.png" /><figcaption>Image credits — Notebooklm</figcaption></figure><p>The system inherently emphasized recent actions over distant ones, since the capacitor charge decayed with time. This temporal credit assignment problem, as we’d call it today, was solved through the physics of the hardware itself. The machine didn’t need to explicitly reason about which actions led to success; the fading capacitor charge automatically gave more weight to the final steps before reaching the goal.</p><h3><strong>The Legacy of 300 Vacuum Tubes</strong></h3><p>SNARC had its limitations, of course. The system required 300 vacuum tubes wired in series — a maintenance nightmare in an era when vacuum tubes failed regularly. Scaling beyond 40 neurons would have been prohibitively expensive and unreliable.</p><p>Yet despite these constraints, SNARC demonstrated something profound: that machines could exhibit adaptive behavior through probabilistic feedback. Nearly three-quarters of a century before modern deep learning, Minsky and Edmonds had built a system that learned from experience rather than following explicit instructions.</p><p>More importantly, SNARC pointed toward a solution to what would later be called the von Neumann bottleneck — the fundamental limitation of architectures where processing and memory are separated, forcing data to shuttle back and forth through a narrow channel. By embedding memory directly into the computational elements themselves, SNARC foreshadowed modern neuromorphic computing, where computation and storage are unified in ways that more closely resemble biological brains.</p><h3><strong>From Mazes to Modern AI</strong></h3><p>Today’s neuromorphic chips, with their millions of artificial neurons and synapses, bear little physical resemblance to that jury-rigged assemblage of bomber parts and vacuum tubes. But they inherit SNARC’s central insight: that intelligence might emerge not from faster symbol manipulation, but from networks of simple elements that adjust their connections based on experience.</p><p>When Minsky built SNARC in 1951, he was trying to understand the human mind by recreating it. In the process, he helped establish the technical feasibility of learning machines and laid conceptual groundwork that continues to shape artificial intelligence seven decades later.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*a3BDt1pq6_toJyprmQitPw.png" /><figcaption>Image credits — Notebooklm</figcaption></figure><p>The profound mystery that motivated Minsky — how does the mind work? — remains unsolved. But SNARC proved that at least some aspects of intelligence could be captured in hardware, learned rather than programmed, and studied as physical phenomena rather than ineffable mysteries.</p><p>Not bad for a graduate project built from salvaged bomber parts.</p><p><a href="https://pubmed.ncbi.nlm.nih.gov/31550405/">Bridging Biological and Artificial Neural Networks with Emerging Neuromorphic Devices: Fundamentals, Progress, and Challenges - PubMed</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a5ddb6daddcc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Thoughts Become Words: A Brain Implant That Restored Speech]]></title>
            <link>https://medium.com/@tharushi.cnnp/when-thoughts-become-words-a-brain-implant-that-restored-speech-f138276e61f5?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/f138276e61f5</guid>
            <category><![CDATA[neurotechnology]]></category>
            <category><![CDATA[bci]]></category>
            <category><![CDATA[als]]></category>
            <category><![CDATA[neuroprosthetics]]></category>
            <category><![CDATA[medical-innovation]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Fri, 14 Nov 2025 08:05:55 GMT</pubDate>
            <atom:updated>2025-11-14T08:15:02.019Z</atom:updated>
            <content:encoded><![CDATA[<p><em>How scientists helped a man with ALS communicate at near-human accuracy using only his brain signals</em></p><p>Imagine trying to speak but being trapped inside your own body, your thoughts clear but your voice stolen by disease. For millions living with conditions like ALS, this isn’t imagination — it’s daily reality.</p><p>But what if we could bypass damaged muscles entirely and read speech directly from the brain?</p><h3>The Clinical Context</h3><p>Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disease that systematically destroys motor neurons — the nerve cells responsible for controlling voluntary muscle movement. As the disease advances, patients lose the ability to walk, use their hands, and eventually speak, while typically retaining full cognitive function.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/685/1*_coWfbLOIvCH6rTMsewEBg.png" /><figcaption><strong>Pathogenesis </strong>-Image credits:<a href="https://www.nature.com/articles/s41573-022-00612-2">https://www.nature.com/articles/s41573-022-00612-2</a></figcaption></figure><p>The participant in this UC Davis study was a 45-year-old left-handed man, five years into his ALS diagnosis at the time of enrollment. His condition had progressed to the point where he was non-ambulatory and required assistance with wheelchair control to personal hygiene. Most significantly, he had developed severe dysarthria — a motor speech disorder that rendered his speech monotone, low-volume, and nasal in quality.</p><p>While his regular care partner could still interpret his speech, it was unintelligible to unfamiliar listeners. This profound communication barrier meant that despite having intact cognitive abilities, he had effectively lost his primary means of expressing thoughts, needs, and personality to most of the world.</p><h3>Reading the Brain’s Speech Centers</h3><p>The UC Davis research team took a bold approach: implant four tiny electrode arrays directly into the left precentral gyrus — a critical brain region that coordinates speech movements. But this wasn’t guesswork. The precision behind the placement was remarkable.</p><h4>Surgical Planning: Mapping the Brain</h4><p>Before surgery, the team used functional MRI (fMRI) to confirm the participant was left-hemisphere dominant for language — crucial information since he was left-handed. They then employed the Human Connectome Project’s high-resolution brain mapping, precisely calibrated to his individual brain anatomy.</p><p>The four arrays were strategically positioned across complementary speech regions:</p><ol><li><strong>Area 55b</strong> — A language-related region involved in phonological representation (how speech sounds are processed)</li><li><strong>Ventral premotor cortex (dorsal aspect)</strong> — Critical for planning and coordinating speech movements</li><li><strong>Ventral premotor cortex (ventral aspect)</strong> — Another speech planning area</li><li><strong>Primary motor cortex (Brodmann area 4)</strong> — The final command centre for executing movements</li></ol><p>This multi-site approach was intentional. Previous studies had shown that the ventral premotor cortex provided strong speech signals, but by adding primary motor cortex and area 55b — regions rarely recorded with such density — the team could capture a more complete picture of speech production.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u9oAF9A1yHhpj6FToqMaQQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TTy4htFENIcZNdAVuPtLmg.png" /><figcaption><strong><em>Electrode Locations and Speech-Decoding Setup.</em></strong><em> — Image credits — </em><a href="https://www.nejm.org/doi/full/10.1056/NEJMoa2314132#v1">https://www.nejm.org/doi/full/10.1056/NEJMoa2314132#v1</a></figcaption></figure><h4>The Hardware: Microelectrode Arrays</h4><p>Each microelectrode array measures just 3.2mm × 3.2mm — smaller than a postage stamp — with 64 electrodes arranged in an 8×8 grid. During a 5-hour surgery under general anesthesia, surgeons made a 5cm × 5cm opening in the skull on the left side.</p><p>Using a specialized high-speed pneumatic inserter (think of it like a precisely controlled nail gun for the brain), they inserted each electrode 1.5 mm into the cortex. Care was taken to avoid large blood vessels visible on the brain’s surface. Each electrode has a recording site approximately 50 micrometers in diameter — about half the width of a human hair — designed to detect electrical activity from one or just a few neurons.</p><p>The four arrays connect to two percutaneous “pedestals” — connectors that protrude through the skin and are secured to the skull with titanium screws. These pedestals have 128 channels each, providing 256 total recording sites. Reference wires</p><p>placed in the subdural and epidural spaces (between and around the brain’s protective layers) help filter out electrical noise.</p><p>Remarkably, the pedestals use standard HDMI cables to transmit neural data to external computers on a wheeled cart. No exotic connections — just technology you’d use for your home theater, repurposed for reading minds.</p><h3>From Brain Signals to Sentences</h3><p>Here’s where it gets fascinating — and technically complex.</p><h3>The Signal Processing Pipeline</h3><p>The NeuroPort System continuously monitors all 256 electrode sites, capturing the electrical chatter of neurons attempting to coordinate speech. This raw neural data flows through a multi-stage processing pipeline:</p><p><strong>Stage 1: Neural Activity to Phonemes (Every 80 milliseconds)</strong></p><p>Every 80 milliseconds — about the duration of a single syllable — the system analyses the pattern of neural firing across all electrodes. It’s looking for the unique “neural signature” associated with each phoneme (the smallest units of speech sound, like “buh,” “aah,” “tuh”).</p><p>Think of it like reading sheet music: different combinations of notes create different sounds. Similarly, different patterns of neural activity across the 256 recording sites indicate different phonemes. A machine learning model trained on the participant’s data learned to recognise these patterns with remarkable accuracy.</p><p>Importantly, <strong>no microphone was used</strong>. The researchers verified there was no acoustic or vibration contamination in the signals. They were reading pure neural intention, not detecting actual sound or vocal cord vibrations.</p><p><strong>Stage 2: Phonemes to Words</strong></p><p>Phonemes alone would produce gibberish. The system uses an openly available language model to convert phoneme sequences into actual English words. This is where probabilistic magic happens — the model knows that certain phoneme combinations are more likely to form real words than others.</p><p>For example, if the neural signal suggests “k-ae-t,” the language model strongly favours “cat” over nonsense combinations.</p><p><strong>Stage 3: Words to Coherent Sentences</strong></p><p>Two additional open-source language models act as the final refinement layer. They analyze the predicted word sequence and translate it into the most likely intended sentence, accounting for grammar, context, and natural language patterns.</p><p>For instance, if the decoded words are “I want water please,” the language models help ensure proper sentence structure and can even add punctuation (a feature added in later sessions).</p><h3>Continuous Calibration: The System Gets Smarter</h3><p>One of this study’s breakthroughs was the continuous calibration approach. Unlike previous systems that required massive upfront training, this decoder improved incrementally.</p><p>Data from multiple sessions were combined and fed back into the system. As the participant used the neuroprosthesis over 32 weeks across 84 sessions, the decoder learned:</p><ul><li>How his individual neurons represented different phonemes</li><li>How his neural patterns varied across different speaking conditions</li><li>Which electrode sites provided the most reliable information</li></ul><h3>Two Evaluation Metrics</h3><p>The researchers measured performance using industry-standard metrics:</p><p><strong>Phoneme Error Rate</strong>: How accurately the system converted neural activity into phonemes <em>before</em> applying language models. This measures the “raw” neural decoding ability.</p><p><strong>Word Error Rate</strong>: The overall communication accuracy after all language models have been applied. This is calculated as the ratio of errors (insertions, deletions, or substitutions needed) to total expected words, expressed as a percentage.</p><p>A 2.5% word error rate means that in every 100 words attempted, only 2.5 required correction to match the intended sentence. That’s astonishingly accurate for direct brain-to-text communication.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fe5A8xEXhnXrjx0Ll7whiw.jpeg" /><figcaption><strong><em>Online Speech-Decoding Performance</em></strong><em>. — Image credits — </em><a href="https://www.nejm.org/doi/full/10.1056/NEJMoa2314132">https://www.nejm.org/doi/full/10.1056/NEJMoa2314132</a></figcaption></figure><h3>Rapid Progress, Remarkable Accuracy</h3><p>The learning curve was extraordinary, especially when compared to previous brain-computer interface attempts.</p><h3>Timeline of Achievement</h3><p><strong>Session 1 (Day 25 post-implantation)</strong>: First closed-loop operation with a 50-word vocabulary. The system worked well enough to bring the participant and his family to tears.</p><p><strong>Session 2</strong>: Vocabulary exploded from 50 words to over 125,000 words — essentially the entire English language. After just 260 additional training sentences collected over 1.4 hours, the system achieved 9.8% word error rate (90.2% accuracy) on copy tasks.</p><p><strong>Session 15</strong>: Error rate dropped to just 2.5% (97.5% accuracy). Total cumulative training time: approximately 16 hours.</p><p><strong>Sessions 80–84</strong>: Maintained 97.5% accuracy with a mean word error rate of 2.5% at the participant’s self-paced speaking rate of 31.6 words per minute. Individual daily word error rates ranged from 1.0% to 3.3%.</p><h3>Putting Accuracy in Context</h3><p>To understand how remarkable this is:</p><ul><li><strong>Professional speech recognition</strong> (like smartphone dictation): ~5% word error rate</li><li><strong>This system</strong>: 2.5% word error rate after 16 cumulative hours of training</li></ul><p>The neuroprosthesis approached human-level accuracy for spontaneous speech decoding. During conversation mode across selected sessions, the mean word error rate was 3.7% — still remarkably low for unstructured, participant-initiated communication.</p><h3>Two Operating Modes</h3><p>Each session consisted of task blocks lasting 5–30 minutes, with breaks in between. The participant worked in two distinct modes:</p><p><strong>Instructed-Delay Copy Task</strong>: Words appeared on screen, and after a visual or audio cue, the participant attempted to say them. This provided known “ground truth” for measuring accuracy and collecting training data. It’s like practicing scales on a piano — not creative, but essential for calibration.</p><p><strong>Participant-Paced Conversation Mode</strong>: The real magic happened here. The participant could say whatever he wanted in unstructured conversation. The system had access to a 125,000-word dictionary (covering most of English) and could:</p><ul><li>Detect when he started or stopped speaking (false triggers occurred less than 1% of the time)</li><li>Display decoded text in real-time as he attempted to speak</li><li>Read completed sentences aloud using text-to-speech</li><li>Automatically add punctuation (in later sessions)</li><li>Function as a Bluetooth keyboard to send text to his personal computer for emails and other tasks</li></ul><p>When the decoder struggled with words outside its vocabulary (like unusual proper nouns), he had a backup: an eye-tracker let him spell words letter-by-letter by attempting to say individual letters. He could also use the eye-tracker to indicate whether decoded sentences were “100% correct,” “mostly correct,” or “incorrect.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mQfS89CNCbhvXXFHBaUv6A.jpeg" /><figcaption><strong><em>Conversation-Mode User Interface. </em></strong><em>— image credits — </em><a href="https://www.nejm.org/doi/full/10.1056/NEJMoa2314132">https://www.nejm.org/doi/full/10.1056/NEJMoa2314132</a></figcaption></figure><h3>What This Means for the Future</h3><p>This study represents a major leap forward in brain-computer interface technology. The rapid calibration, high accuracy, and extensive vocabulary demonstrate that restoring communication for people with paralysis isn’t just theoretically possible — it’s practically achievable.</p><h3>The Bigger Picture</h3><p>We’re entering an era where the boundary between brain and machine is becoming increasingly permeable. Not in a dystopian sense, but in a profoundly human one — giving voice to the voiceless, restoring agency to those who’ve lost it.</p><p>For the millions of people worldwide who’ve lost the ability to speak due to ALS, stroke, brain injury, or other conditions, research like this demonstrates tangible progress.</p><p>Progress showing that thoughts don’t have to stay trapped inside.</p><p>Progress proving that technology can bridge the gap between intention and expression.</p><p>Progress indicating that one day, losing your voice won’t mean losing your ability to be heard.</p><p><em>Based on research conducted by UC Davis and published in The New England Journal of Medicine. The study represents collaborative work in neuroprosthetics, brain-computer interfaces, and assistive technology.</em></p><p>#BCI #Neuroprosthetics #NeuralEngineering #Neurotechnology #MedicalInnovation #RealtimeDecoding #FutureOfMedicine</p><p><a href="https://www.nejm.org/doi/full/10.1056/NEJMoa2314132">https://www.nejm.org/doi/full/10.1056/NEJMoa2314132</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f138276e61f5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Era of Multi-Brain Communication: How BrainNet is Revolutionizing Collaborative BCIs]]></title>
            <link>https://medium.com/@tharushi.cnnp/the-era-of-multi-brain-communication-how-brainnet-is-revolutionizing-collaborative-bcis-279af3f6192a?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/279af3f6192a</guid>
            <category><![CDATA[neurotechnology]]></category>
            <category><![CDATA[eeg]]></category>
            <category><![CDATA[brain-computer-interface]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Sun, 02 Nov 2025 13:18:52 GMT</pubDate>
            <atom:updated>2025-11-02T13:18:52.168Z</atom:updated>
            <content:encoded><![CDATA[<p>For decades, brain-computer interfaces (BCIs) operated on a simple premise: one brain, one machine. But what if multiple brains could communicate directly with each other, forming networks of minds that solve problems together? This is the promise of multi-person brain-to-brain interface systems, and researchers at the University of Washington and Carnegie Mellon University have proven it’s not just possible — it’s remarkably effective.</p><h3>From Single-Brain to Multi-Brain Systems</h3><p>Traditional BCIs focused on connecting one person’s brain to a computer or device. While revolutionary for helping paralyzed individuals control prosthetics or type with their thoughts, these systems were fundamentally limited to one-way, single-user interactions.</p><p>Multi-brain BCI systems represent an entirely new paradigm. Instead of brain-to-machine communication, we’re now entering the era of brain-to-brain networks where multiple people can collaborate and exchange information directly through neural signals — no words, no gestures, just pure thought transmission.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5Ol0chqB58Jnzjh70EyJXg.png" /></figure><h3>BrainNet: The First Multi-Person Brain Interface</h3><p>The breakthrough came with BrainNet, the first multi-person non-invasive direct brain-to-brain interface for collaborative problem-solving. The system demonstrates what happens when you network human minds together like computers on the internet.</p><p>BrainNet combines two established neurotechnologies:</p><p><strong>Electroencephalography (EEG)</strong> to record brain signals from the scalp, capturing the electrical patterns of neural activity.</p><p><strong>Trans-cranial Magnetic Stimulation (TMS)</strong> to deliver information non-invasively to the brain through carefully calibrated magnetic pulses.</p><p>But the real innovation isn’t the individual technologies — it’s how they’re orchestrated to create a functioning network of three human brains working as one problem-solving unit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UMMzyOnKtIth4vUZM4leNQ.png" /><figcaption><strong>Architecture of BrainNet</strong>. Image Credits — <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC6467884/#Sec1">https://pmc.ncbi.nlm.nih.gov/articles/PMC6467884/#Sec1</a></figcaption></figure><h3>How Multi-Brain Collaboration Actually Works</h3><p>The researchers designed an elegant experiment using a Tetris-like game that required genuine collaboration between three participants:</p><p><strong>The Network Architecture:</strong></p><ul><li><strong>Two “Sender” brains</strong> observe the complete game state and make independent decisions</li><li><strong>One “Receiver” brain</strong> integrates signals from both Senders without seeing the full picture</li><li>All three communicate through Internet-connected brain interfaces</li></ul><p><strong>The Communication Protocol:</strong></p><p>Senders make decisions by focusing attention on LED lights flashing at specific frequencies (17 Hz for “rotate the block,” 15 Hz for “don’t rotate”). Their brain’s electrical response to these frequencies is decoded in real-time by EEG sensors, extracting their decision.</p><p>These decisions are transmitted over the Internet to the Receiver’s location, where they’re converted into TMS pulses delivered to the occipital cortex. A pulse above threshold creates a phosphene — a flash of light perceived in the visual field — signalling “yes.” Below threshold means “no.”</p><p>The Receiver’s brain integrates these two independent signals, makes a decision, and responds using their own EEG interface.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/749/1*NxEEanIyDkttSY6fCAu4Qg.jpeg" /><figcaption><strong>How the game works</strong> — Image Credits — <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC6467884/#Sec1">https://pmc.ncbi.nlm.nih.gov/articles/PMC6467884/#Sec1</a></figcaption></figure><p><strong>The Innovation: Two Rounds of Interaction</strong></p><p>Unlike previous brain-to-brain interfaces that were one-directional, BrainNet implements bidirectional communication. After the first round, Senders see the Receiver’s decision and can send corrective signals if needed. This creates a true feedback loop — a conversation between brains.</p><h3>Networks Learn Like Social Systems</h3><p>Five groups of three people each used BrainNet to complete 16 collaborative trials. The results were striking:</p><p><strong>Average accuracy: 81.25%</strong> in correctly solving the task through pure brain-to-brain communication.</p><p>But the most fascinating discovery came when researchers introduced noise into one Sender’s signals, making them unreliable in 10 out of 16 trials. Would the Receiver’s brain learn to distinguish reliable from unreliable information sources?</p><p>It did. Just like we learn which friends give better advice in social networks, Receivers unconsciously learned to weigh information from the reliable Sender more heavily. The mutual information analysis showed significantly higher information transfer between Receivers and good Senders compared to bad ones — and this learning happened over the course of just 16 trials.</p><p>The brain’s ability to evaluate source reliability purely from transmitted neural signals suggests that multi-brain BCI networks could scale to larger groups, with individual brains naturally learning which connections to trust.</p><h3>Beyond Three Brains: Scaling Neural Networks</h3><p>BrainNet’s architecture isn’t limited to three participants. The system can theoretically scale to include larger numbers of Senders, all transmitting information to one or more Receivers. Imagine:</p><ul><li><strong>Distributed decision-making:</strong> Multiple experts sending their analysis directly to a decision-maker’s brain</li><li><strong>Collective problem-solving:</strong> Teams of minds attacking complex challenges from different angles simultaneously</li><li><strong>Global brain networks:</strong> Cloud-based BBI servers directing information between any connected brains worldwide</li></ul><p>The researchers envision a future where brain-to-brain interfaces create true “social networks of brains” — not metaphorically, but literally.</p><h3>Medical and Therapeutic Applications</h3><p>Brain-to-brain communication systems are showing enormous promise for medical rehabilitation and therapy:</p><p><strong>Locked-In Syndrome and Communication Disorders:</strong> Multi-brain BCI systems equipped with brain-to-brain coupling technology could help doctors communicate with patients who cannot speak or move, creating communication channels when all others have failed.</p><p><strong>Stroke and Neurological Recovery:</strong> Research indicates these systems can improve recovery rates by harnessing natural human social connections. The bidirectional nature of neural coupling might help damaged brains relearn lost functions by directly interfacing with healthy neural patterns.</p><p><strong>Cognitive Rehabilitation:</strong> Direct brain-to-brain interfaces could accelerate learning and rehabilitation by allowing therapists to transmit not just instructions, but actual neural patterns associated with proper motor control or cognitive function.</p><p><strong>Rapid Knowledge Transfer:</strong> For healthy individuals, future multi-brain BCIs could create new methods for rapid idea sharing — imagine directly experiencing another person’s insight or understanding a complex concept through neural transmission rather than hours of explanation.</p><h3>A New Frontier in Human Connection</h3><p>BrainNet and emerging multi-brain BCI systems represent more than technological achievement — they’re fundamentally expanding what it means to communicate, collaborate, and connect as humans.</p><p>We’re moving from an era where brain-computer interfaces helped individuals control machines, to one where networks of brains can merge their cognitive resources into something greater than the sum of their parts.</p><p>The implications stretch from medical miracles — giving voice to the voiceless — to transforming how humanity tackles its greatest challenges through genuine collective intelligence.</p><p>As these technologies mature within ethically-grounded frameworks, we’re not just building better communication tools. We’re engineering the first true social networks of minds, where “I’ll share my thoughts with you” becomes literal reality.</p><p>The future of BCIs isn’t singular — it’s plural, networked, and collaborative. It’s multi-brain systems working together, and that future has already begun.</p><p><em>Based on research published in Scientific Reports: “BrainNet: A Multi-Person Brain-to-Brain Interface for Direct Collaboration Between Brains” by Jiang et al., 2019.</em></p><p><a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC6467884/#Sec1">BrainNet: A Multi-Person Brain-to-Brain Interface for Direct Collaboration Between Brains</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=279af3f6192a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Future is Already Here: Why I’m Obsessed with taVNS (And You Should Be Too)]]></title>
            <link>https://medium.com/@tharushi.cnnp/the-future-is-already-here-why-im-obsessed-with-tavns-and-you-should-be-too-dd8ae67747d8?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/dd8ae67747d8</guid>
            <category><![CDATA[personalized-medicine]]></category>
            <category><![CDATA[ai-in-healthcare]]></category>
            <category><![CDATA[neurotechnology]]></category>
            <category><![CDATA[biotech]]></category>
            <category><![CDATA[neuromodulation]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Sat, 04 Oct 2025 06:24:55 GMT</pubDate>
            <atom:updated>2025-11-05T15:47:01.508Z</atom:updated>
            <content:encoded><![CDATA[<h3>Transcutaneous auricular Vagus Nerve Stimulation</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*grtoo03SiB65I4Ji3A0Y7w.png" /></figure><p><em>How a small electrical pulse on your ear is bridging neuroscience, bioelectronics, and the future of personalised medicine</em></p><p>To be honest — when I first stumbled upon Transcutaneous auricular Vagus Nerve Stimulation (taVNS) while deep-diving into neuro-modulation research, I thought it was almost too elegant to be true. A non-invasive device that could potentially treat depression, aid stroke recovery, and even tackle obesity? By stimulating a nerve in your <em>ear</em>?</p><p>Let me break down why taVNS represents everything I believe about the intersection of biotech, AI, and medicine.</p><h3>The Biological Interface We’ve Been Overlooking</h3><p>Here’s something wild: your outer ear — specifically the <strong>cymba concha</strong> and <strong>cavum concha</strong> — is directly wired to the vagus nerve, one of the most influential communication channels in your body. This nerve is like your body’s central nervous system hotline, connecting your brain to virtually every major organ.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*moXKq9DpcjaSXuN1OKONLw.jpeg" /><figcaption>Source:<a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC12174914/#s0003">https://pmc.ncbi.nlm.nih.gov/articles/PMC12174914/#s0003</a>, AVBN — Auricular Branch of Vagus Nerve</figcaption></figure><p>For years, we’ve known that stimulating the vagus nerve could treat conditions like epilepsy and treatment-resistant depression. But traditional vagus nerve stimulation (VNS) required surgery — a device implanted in your chest with wires running to your neck. Effective, sure, but hardly accessible or scalable.</p><p>taVNS changes the game entirely. By targeting that auricular branch through the skin of the ear, we get the therapeutic benefits without the surgical barrier. No implants. No operating rooms. Just bioelectronic precision delivered non-invasively.</p><p>This is the kind of innovation that gets me excited about biotech — taking fundamental biological insights and making them accessible to millions.</p><h3>The Neuroscience is Mind-Blowing — Vagal Cortical Pathway Model</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MgsY2bnzviq9GhqK8H57ow.jpeg" /><figcaption>Source: <a href="https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2024.1527842/full">https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2024.1527842/full</a></figcaption></figure><p>When that tiny electrical impulse hits the auricular vagus nerve, it’s like pressing a master switch that cascades through your entire brain. The signal reaches the nucleus tractus solitarius (NTS) in your brainstem, which then activates a constellation of critical neural hubs:</p><p>🧠 <strong>The Locus Coeruleus</strong> releases norepinephrine across the cortex — essential for attention, arousal, and neuroplasticity (the brain’s ability to rewire itself)</p><p>🧠 <strong>The Dorsal Raphe Nuclei</strong> modulate serotonin pathways, influencing mood and how different brain networks coordinate</p><p>🧠 <strong>Cholinergic systems</strong> pump out acetylcholine, boosting cognition and arousal</p><p>But here’s what really blows my mind: taVNS significantly upregulates <strong>brain-derived neurotrophic factor</strong> (BDNF) — basically growth fertilizer for your neurons. BDNF activates molecular cascades that are fundamental to learning, memory formation, and synaptic plasticity. We’re not just treating symptoms; we’re enhancing the brain’s capacity to heal and adapt at the cellular level.</p><p>This is precision neuro-modulation at its finest.</p><h3>Real-World Impact: Where Science Meets Humanity</h3><h4>Treatment-Resistant Depression: A New Hope</h4><p>About one-third of people with major depressive disorder don’t respond adequately to antidepressants. For them, taVNS offers a non-pharmacological alternative that actually works.</p><p>Neuroimaging studies show that taVNS immediately regulates neural synchrony in the medial orbitofrontal cortex and enhances emotion regulation circuits. It reduces the processing of negative emotions by modulating the Default Mode Network, frontoparietal control network, and reward pathways. Even more remarkably, it normalizes abnormal connectivity patterns between brain regions that depression disrupts.</p><p>This isn’t just about treating symptoms — it’s about restoring healthy brain communication patterns.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uusXMQncZBQ07SICqj65TQ.png" /></figure><h4>Obesity: Hacking Appetite at the Neural Level</h4><p>Obesity is a complex metabolic and neural disorder, not a failure of willpower. taVNS addresses it at the root by modulating orexin-dependent circuits in the hypothalamus — the brain’s command center for hunger, satiety, and energy balance.</p><p>By enhancing insulin sensitivity and suppressing appetite through GABAergic projections that regulate orexin-producing neurons, taVNS offers a neurobiological intervention for a problem that affects over 40% of adults in many developed countries.</p><p>This is the kind of application that could genuinely transform public health.</p><h3>Where AI Meets Biotech: Closed-Loop Systems</h3><p>Here’s where things get <em>really</em> exciting for someone like me who’s fascinated by the convergence of AI, biotech, and medicine.</p><h4>Motor-Activated Auricular Vagus Nerve Stimulation (MAAVNS)</h4><p>This brilliant closed-loop system uses surface EMG to detect muscle movements and delivers time-locked taVNS within 100 milliseconds of detected motor activity. For stroke patients in rehabilitation, this synchronization of neuro-modulation with voluntary movement produces significantly better motor recovery than continuous stimulation.</p><p>Why? Because the brain learns best when the modulatory signal arrives exactly when the relevant neural circuits are active. It’s like having a smart trainer that knows the perfect moment to amplify your effort.</p><h4>ECG-Gated taVNS: Cardiac-Synchronized Neuro-modulation</h4><p>Imagine timing brain stimulation to your heartbeat — delivering pulses during systole to maximize vagal activation that naturally aligns with cardiovascular rhythms. Early studies show this produces better modulation of heart rate variability and autonomic function than untimed stimulation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xuXsEN8jXcKZRIUzqE03yQ.png" /></figure><p>It’s biomimetic design at its best — working with the body’s natural rhythms rather than imposing arbitrary stimulation patterns.</p><h3>The Convergence Vision: Multi-Modal Adaptive Therapy</h3><p>The future I’m most excited about? Multi-modal closed-loop systems that simultaneously monitor EEG, cardiac rhythms, respiratory patterns, and motor activity to deliver truly personalized, adaptively optimized neuro-modulation.</p><p>This represents a fundamental paradigm shift:</p><ul><li><strong>From one-size-fits-all to personalized</strong>: Therapy adapts to your unique physiology</li><li><strong>From reactive to predictive</strong>: AI anticipates optimal treatment windows</li><li><strong>From invasive to accessible</strong>: Non-invasive tech democratizes advanced neuroscience</li><li><strong>From symptomatic to systems-based</strong>: We address root causes through biological pathways</li></ul><p>This is the intersection of neuroscience, bioelectronics, machine learning, and human-centered design — a scenary of real 21st-century medicine.</p><h3>The Path Forward</h3><p>We’re still early in the taVNS journey. There’s work to be done on optimizing stimulation parameters, understanding individual response variability, integrating AI for adaptive therapy, and conducting larger clinical trials. But the foundation is solid, and the potential is enormous.</p><p>This is exactly the kind of work that excites me about the future of biotech — where deep biological understanding meets cutting-edge technology to create solutions that are both scientifically sophisticated and practically accessible.</p><p>The vagus nerve has been transmitting signals for as long as humans have existed. We’re just now learning to join the conversation. And I believe that conversation is going to change millions of lives.</p><p><em>What emerging biotechnologies are you excited about? Let’s discuss in the comments — I’d love to hear what’s capturing your imagination at the intersection of AI, medicine, and human potential.</em></p><p>#Neurotechnology #Biotech #DigitalHealth #Neuromodulation #PersonalizedMedicine #AIinHealthcare #Innovation #FutureOfMedicine</p><ul><li><a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC10363288/">Motor Activated Auricular Vagus Nerve Stimulation as a Potential Neuromodulation Approach for Post-Stroke Motor Rehabilitation: A Pilot Study</a></li><li><a href="https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2020.00933/full">Frontiers | Transcutaneous Auricular Vagal Nerve Stimulation and Disorders of Consciousness: A Hypothesis for Mechanisms of Action</a></li><li><a href="https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2024.1527842/full">Frontiers | Mechanisms of vagus nerve stimulation for the treatment of neurodevelopmental disorders: a focus on microglia and neuroinflammation</a></li><li><a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC12174914/#s0003">Transcutaneous Auricular Vagus Nerve Stimulation (taVNS) for Insomnia Disorder: A Narrative Review of Effectiveness, Mechanisms and Recommendations for Clinical Practice</a></li><li><a href="https://pubmed.ncbi.nlm.nih.gov/40456821/">Transcutaneous auricular vagus nerve stimulation inhibits food intake and body weight gain through the orexin dependent pathway in high fat diet mice - PubMed</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dd8ae67747d8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Information That Changes Everything: The Science of System Control]]></title>
            <link>https://medium.com/@tharushi.cnnp/information-that-changes-everything-the-science-of-system-control-b5c23d77d5fb?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/b5c23d77d5fb</guid>
            <category><![CDATA[systems-thinking]]></category>
            <category><![CDATA[brain-computer-interface]]></category>
            <category><![CDATA[cybernetics]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Sun, 21 Sep 2025 16:14:12 GMT</pubDate>
            <atom:updated>2025-09-21T16:14:12.822Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>systemInformation That Changes Everything: The Science of System Control</strong></h3><p>How a system functions, coordinates between its elements and responds to their environment is via the information flow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*VutkKdSq1_5uVHEWjyGTXg.jpeg" /></figure><h3>Information</h3><p>What actually is information?</p><p>As Gregory Bateson — a contributory cybernetician — puts, information is “a difference that makes a difference”.</p><p>Now, what does this beautiful idea really mean?</p><p>Let’s take a simple example. Your normal body temperature is 37°C. When your body temperature rises due to fever, the temperature sensors in your body sense this difference and inform the brain (hypothalamus — temperature regulating center in the brain) about that. It processes the data and triggers cooling mechanisms like sweating and vasodilation (dilating your blood vessels) to increase heat loss. The increment of body temperature more than the normal value is the information. Its presence made a difference in the system called “your body”.</p><p>When you play chess with a game playing AI, your new move and the current board position is the information the AI gets. Neural networks evaluate millions of possible future moves, assigning probability values to winning chances for each option thus triggering a chance in the system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vq4LcSj_-B0pNE-QeGUZAA.png" /></figure><p>But information alone isn’t enough — what matters is what the system does with it. That’s where control comes in. Let’s dive a bit deeper into the scenario.</p><h3>The Control</h3><p>Your normal body temperature is the desired state of your body. When it increases more than the normal, a “gap” is created between the current state and your body’s desired state. Via simple mechanisms like sweating, your body achieves its desired state from the current undesired state reducing the gap between the two states. This is called “the control” in cybernetics. For control, information flow is essential. And interestingly, this is a system that self-regulates through information processing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*smOf2Z3C03M8g5O5XWU4gQ.png" /></figure><p>Consider controlling a cursor using a BCI (Brain Computer Interface).Its desired state is moving a computer cursor to a specific target on screen (patient’s intended destination). The current state is cursor’s actual position on the display. The BCI system compares intended movement direction (decoded from motor cortex signals) with actual cursor movement. The system translates neural firing patterns into cursor movement commands, adjusting velocity and direction to close the gap. Thus achieves the control.</p><h4>Final thoughts</h4><p>From sweating during a fever to moving a cursor with thought, systems use information to sense, compare, and act. And that’s the essence of control — the cybernetic loop that keeps systems alive, adaptive, and intelligent.</p><p>#SystemsThinking #Cybernetics #InformationTheory #ControlSystems #Neuroscience #ArtificialIntelligence #MachineLearning #BrainComputerInterfaces</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b5c23d77d5fb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why whole systems outsmart their pieces : Emergence]]></title>
            <link>https://medium.com/@tharushi.cnnp/why-whole-systems-outsmart-their-pieces-emergence-8d311a392469?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/8d311a392469</guid>
            <category><![CDATA[systems-thinking]]></category>
            <category><![CDATA[cybernetics]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Thu, 18 Sep 2025 17:42:00 GMT</pubDate>
            <atom:updated>2025-09-19T11:44:18.532Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1010/1*F0koG21d3FE2_Xx4Ls_zkw.jpeg" /></figure><p>Recently, I’ve been fascinated by “Cybernetics” , the field that explores systems and their behaviours. At its core, it meticulously explores the concept of “systems”.</p><h3>What is a system?</h3><p>It’s a group of interrelated elements that interact or work together towards a purpose. The elements here have interactions. For example, take the human brain. It basically has different types of neurons and neurochemicals as elements. They engage in a continuous process of signalling and interacting with each other in the process of transmitting sensory information, processing them, and transmitting different output signals to attain a meaningful outcome.</p><h3>Emergence</h3><p>A fascinating phenomenon that occurs when a system operates is “emergence”. Emergence is not the additive result of a functioning system, but the synergistic effect. It’s an entirely new complex behaviour occurring in a functioning system that is not merely the sum of its elements but more.</p><p>Let’s think of some examples…</p><p>Different organic and inorganic molecules interact with each other and combine to form large, intricate and complex structures called “organisms”. Organisms are living creatures, not just a clump of organic material. But an emergence from the clump of organic materials interacting with each other.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bdjn4d7fnxnykgMv4Yq9PA.jpeg" /></figure><p>Billions of neurons in the human brain engage in a continuous harmony of firing and signalling using neuro-chemicals giving rise to emotions, behaviours and most importantly, consciousness.Consciousness isn’t merely the neurons and their firing but an emergence within.</p><p>Large language models trained for next word prediction showing new abilities like translation and reasoning is an emergence from the dynamics of the model.</p><p>AlphaGo trained to play Go game according to rules, discovering novel strategies that surprised even human experts is the same phenomenon.</p><h4>Final thoughts</h4><p>Whether in neurons, algorithms or even societies, emergence is the hidden force shaping complexity into meaning. And it’s this force that makes systems worth studying — because somewhere in the connections, something greater is always waiting to appear.</p><p>Hoping to dive deeper into the concepts soon…</p><p>#SystemsThinking #Cybernetics #Complexity #Emergence #Systems #Neuroscience #ArtificialIntelligence #MachineLearning #AI #LLM #AlphaGo</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8d311a392469" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Predicting Diabetes with Machine Learning: A Deep Dive into XGBoost for Healthcare]]></title>
            <link>https://medium.com/@tharushi.cnnp/predicting-diabetes-with-machine-learning-a-deep-dive-into-xgboost-for-healthcare-f6782593b2ad?source=rss-de5983df3c8d------2</link>
            <guid isPermaLink="false">https://medium.com/p/f6782593b2ad</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[risk-prediction]]></category>
            <category><![CDATA[ai-in-healthcare]]></category>
            <category><![CDATA[healthcare]]></category>
            <dc:creator><![CDATA[Tharushi  Karunaratne]]></dc:creator>
            <pubDate>Sun, 02 Mar 2025 11:19:30 GMT</pubDate>
            <atom:updated>2025-03-02T11:19:30.157Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Diabetes is a growing global health challenge, affecting millions worldwide. Early detection is crucial in preventing complications and improving patient outcomes. In this project, I built a <strong>machine learning model using XGBoost</strong> to predict diabetes in individuals based on clinical data. The goal was to explore how AI can assist healthcare providers in identifying at-risk patients earlier, leading to better preventive care.</p><h3>Dataset Overview: Pima Indians Diabetes Dataset</h3><p>For this project, I used the <strong>Pima Indians Diabetes dataset</strong>, a well-known dataset in medical machine learning research. It contains <strong>768 samples</strong>, including <strong>8 key health indicators</strong>, such as:</p><ul><li>Glucose Level</li><li>BMI (Body Mass Index)</li><li>Blood Pressure</li><li>Age</li><li>Pregnancy Count</li><li>Skin Thickness</li><li>Insulin Level</li><li>Diabetes Pedigree Function</li></ul><p>The <strong>target variable</strong> (Outcome) is binary:</p><ul><li><strong>1</strong> = Diabetic</li><li><strong>0</strong> = Non-diabetic</li></ul><h3>Model Selection:</h3><p>XGBoost (Extreme Gradient Boosting) is a powerful <strong>ensemble learning algorithm</strong> that excels in structured/tabular data. It offers:<br>✅ <strong>High accuracy</strong> due to boosting techniques<br>✅ <strong>Built-in regularization</strong> to reduce overfitting<br>✅ <strong>Fast training speed</strong> and scalability</p><h3>Preprocessing and Model Training</h3><h4>Step 1: Data Cleaning &amp; Preprocessing</h4><ul><li>Normalized data for better model performance.</li><li>Split the dataset into <strong>80% training</strong> and <strong>20% testing</strong>.</li></ul><h4>Step 2: Model Training with Hyperparameter Tuning</h4><p>I tuned key hyperparameters like:</p><ul><li>learning_rate = 0.05</li><li>max_depth = 3</li><li>n_estimators = 100</li></ul><p>Used <strong>cross-validation</strong> to ensure model generalization.</p><h3>Model Performance and Results</h3><p>After training and evaluation, my model achieved:<br>📌 <strong>Accuracy</strong>: <strong>75%</strong><br>📌 <strong>Precision</strong>: <strong>0.81 (Non-diabetic), 0.65 (Diabetic)</strong><br>📌 <strong>F1 Score</strong>: <strong>0.65 for diabetic cases</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/952/1*K_AgkATCBIU4yPhbwn4qUw.png" /><figcaption>Model Performance Statistics</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/501/1*P-9vvjq8Os8Q7r_UZfM4ug.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/536/1*SF1Z6un0cux1npWyqSIhpQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/536/1*fFxl80dX8CL52K3tYQ83QQ.png" /></figure><h3>Insights:</h3><ul><li>The model performs <strong>well</strong> in predicting non-diabetic individuals.</li><li><strong>Diabetic case detection needs improvement</strong> due to class imbalance.</li></ul><h3>Challenges and Future Improvements</h3><p>💡 <strong>Class Imbalance</strong>: The dataset has <strong>fewer diabetic cases</strong>, leading to lower recall for Class 1.<br>💡 <strong>Alternative Models</strong>: Testing <strong>Random Forest </strong>and <strong>Neural Networks</strong> for improved results.<br>💡 <strong>Oversampling &amp; Undersampling</strong>: Techniques like <strong>SMOTE (Synthetic Minority Over-sampling Technique)</strong> could improve diabetic predictions.</p><h3>Real World Relevance and Clinical Impact</h3><ul><li>Early detection of diabetes <strong>can save lives</strong>.</li><li>AI can assist <strong>clinicians in decision-making</strong>, providing additional insights.</li><li>Predictive models <strong>reduce healthcare costs</strong> by enabling early interventions.</li></ul><h3>Final Thoughts:</h3><p>This project has been an exciting step in applying AI to real-world medical challenges. I look forward to improving the model and exploring its potential.</p><p>If you’re interested in <strong>AI in healthcare</strong>, let’s discuss ways to refine and apply machine learning for <strong>better patient outcomes</strong>! Feedback and suggestions are welcome.</p><p>#MachineLearning #AI #Healthcare #DiabetesPrediction #XGBoost #DataScience #ClinicalAI #Neurotech</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f6782593b2ad" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>