<?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 Susan Ruyu Qi on Medium]]></title>
        <description><![CDATA[Stories by Susan Ruyu Qi on Medium]]></description>
        <link>https://medium.com/@susan.ophtho?source=rss-fdbfaaae6aa8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*DzKroSqDmIj--w-GuMm8gA.jpeg</url>
            <title>Stories by Susan Ruyu Qi on Medium</title>
            <link>https://medium.com/@susan.ophtho?source=rss-fdbfaaae6aa8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 18:14:21 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@susan.ophtho/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[Google’s AI can see through your eyes what doctors can’t]]></title>
            <link>https://medium.com/health-ai/googles-ai-can-see-through-your-eyes-what-doctors-can-t-c1031c0b3df4?source=rss-fdbfaaae6aa8------2</link>
            <guid isPermaLink="false">https://medium.com/p/c1031c0b3df4</guid>
            <category><![CDATA[health]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[digital-health]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[tech]]></category>
            <dc:creator><![CDATA[Susan Ruyu Qi]]></dc:creator>
            <pubDate>Mon, 26 Nov 2018 18:33:49 GMT</pubDate>
            <atom:updated>2018-11-27T03:20:33.393Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*_afrr_hKZMg-hMxVLtHnHg.gif" /></figure><p><strong>Can you tell from the photo below if the patient is male or female?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/250/0*1aDzQZiGJRZtpvz3.jpg" /><figcaption>Google’s AI can tell gender by simply looking at a retinal photo like this one</figcaption></figure><p>Human ophthalmologists can only take a guess, and there is a 50:50 chance of getting it right or wrong.</p><h4>Google’s AI, however, can predict it very well (AUC 0.97).</h4><p><em>How is it possible? </em>We don’t know yet but these results are definitely striking. Perhaps the neural network picked up patterns that are too subtle to the human eye.</p><p>To better understand this AI “black box”, a deep-learning technique called <strong>soft attention </strong>was used to identify the anatomical regions that are most important for the model to generate these gender predictions.</p><p><strong>Attention maps (saliency or heat maps)</strong> were generated, representing what the model was“looking at” in each fundus photo in order to predict gender. These maps were then given to 3 ophthalmologists to identify the highlighted features in these maps. Ophthalmologists were blinded to the output predictions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/551/1*LDD9JksFtEoaBsZNoZJANQ.png" /><figcaption>soft attention heat map is overlain in green, indicating the areas of the heat map that the neural-network model is using to make the gender prediction for the image.</figcaption></figure><p><strong>100 heat maps</strong> were randomly selected and interpreted by ophthalmologists. They found that 71% of the heat maps highlighted <strong>vessels;</strong> 78% of the maps highlighted the <strong>optic disc</strong>, and 50% of the maps also highlighted<strong> non-specific features</strong>. <em>In other words, the neural network was looking at a bit of everything but nothing in particular…</em></p><p>Until this day, we didn’t think that there were any differences between male and female eyes. How can a neural network tell them apart from just a photo of the retina is yet to be elucidated.</p><p>Although predicting gender from a retinal photo isn’t directly useful in patient care, this discovery will guide future research to investigate this gender difference, previously unknown to us. We can be hopeful that a better understanding of the anatomy and physiology differences between the male and female eye will lead to a better understanding of eye diseases and in turn benefit drug discoveries.</p><h3>Read more from Health.AI:</h3><p><a href="https://medium.com/health-ai/google-deepmind-might-have-just-solved-the-black-box-problem-in-medical-ai-3ed8bc21f636">Google DeepMind might have just solved the “Black Box” problem in medical AI</a></p><p><a href="https://medium.com/health-ai/ai-2-0-in-ophthalmology-googles-second-publication-c3b5390c19ae">AI 2.0 in Ophthalmology — Google’s Second Publication</a></p><p><a href="https://medium.com/health-ai/deep-learning-in-ophthalmology-using-128-175-retinal-images-59814e8a3f68">Deep Learning in Ophthalmology — How Google Did It</a></p><p><a href="https://medium.com/health-ai/machine-learning-and-oct-images-the-future-of-ophthalmology-47dc64ee9dc6">Machine Learning and OCT Images — the Future of Ophthalmology</a></p><p><a href="https://medium.com/health-ai/machine-learning-and-plastic-surgery-95a8d75a631d">Machine Learning and Plastic Surgery</a></p><p><a href="https://medium.com/health-ai/ai-neural-network-for-pediatric-cataract-6a4341f31263">AI &amp; Neural Network for Pediatric Cataract</a></p><p><a href="https://medium.com/health-ai/the-cutting-edge-the-future-of-surgery-9dfa4315ae4f">The Cutting Edge: The Future of Surgery</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c1031c0b3df4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/health-ai/googles-ai-can-see-through-your-eyes-what-doctors-can-t-c1031c0b3df4">Google’s AI can see through your eyes what doctors can’t</a> was originally published in <a href="https://medium.com/health-ai">Health.AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google DeepMind -Improving the interpretability of “Black Box” in medical imaging]]></title>
            <link>https://medium.com/health-ai/google-deepmind-might-have-just-solved-the-black-box-problem-in-medical-ai-3ed8bc21f636?source=rss-fdbfaaae6aa8------2</link>
            <guid isPermaLink="false">https://medium.com/p/3ed8bc21f636</guid>
            <category><![CDATA[medicine]]></category>
            <category><![CDATA[health-technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ophthalmology]]></category>
            <dc:creator><![CDATA[Susan Ruyu Qi]]></dc:creator>
            <pubDate>Thu, 23 Aug 2018 04:23:35 GMT</pubDate>
            <atom:updated>2021-04-12T01:32:34.057Z</atom:updated>
            <content:encoded><![CDATA[<figure><a href="https://deepmind.com/blog/moorfields-major-milestone/"><img alt="" src="https://cdn-images-1.medium.com/max/763/1*cnyoA2T8BFZRBYWnUlYEtQ.gif" /></a><figcaption>Segmentation Video of an OCT Scan/ taken from DeepMind’s original publication</figcaption></figure><blockquote>The key barrier for AI in healthcare is the “Black Box” problem. For most AI systems, the model is hard to interpret and it is difficult to understand why they make a certain diagnosis or recommendation. This is a huge issue in medicine, for both physicians and patients.</blockquote><p>Deep Mind’s study published last week in <a href="https://www.nature.com/articles/s41591-018-0107-6">Nature Medicine</a>, presenting their Artificial Intelligence (AI) product capable of diagnosing many ophthalmic conditions from 3D retinal OCT scans. Its performance is on par with the best retinal specialists and superior to some human experts.</p><p>This AI product’s accuracy and range of diagnoses are certainly impressive. It is also the first AI model to reach expert-level performance with 3D diagnostic scans. From a clinical point of view, however, what is even more groundbreaking is the ingenious way in which this AI system operates and mimics the real-life clinical decision process. It addresses the “Black Box” issue which has been one of the biggest barriers to the integration of AI technologies in healthcare.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*b0wGucm_EhTVE88PmhWGlQ.gif" /><figcaption>An optical coherence tomography (OCT) scan of the retina</figcaption></figure><h3>Two Neural Networks</h3><p>DeepMind’s AI system improved the interpretability of the “Black Box” by creating a framework with two separate neural networks. Instead of training one single neural network to identify pathologies from medical images, which would require a lot of labelled data per pathology, their framework decouples the process into two: 1) <strong>Segmentation</strong> (identify structures on the images) 2) <strong>Classification</strong> (analyze the segmentation and come up with diagnoses and referral suggestions)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/902/1*wiqWqvcMLBr82qZ1_RqmrA.png" /><figcaption>DeepMind’s framework tackles the “Black Box” Problem by having 2 neural networks with a readily viewable intermediate representation (tissue map) in between</figcaption></figure><h4>1. The Segmentation Network</h4><p>Using a three-dimensional U-Net architecture, this first neural network translates raw OCT scans into tissue maps. It was trained using 877 clinical OCT scans. For each scan’s 128 slices, only about 3 representative ones were manually segmented. This sparse annotation procedure significantly reduced workload and allowed them to cover a large variety of scans and pathologies. The tissue maps identify the shown anatomy (ten layers of the retina) and label disease features (intra-retinal fluid, hemorrhage) and artifacts.</p><p>This process mimics the typical clinical decision process. It allows physicians to inspect the AI’s segmentation and gain insight into the neural network’s “reasoning”. This<em> intermediate representation</em> is key to the future integration of AI into clinical practice. It is particularly useful in difficult and ambiguous cases. Physicians can inspect and visualize the automated segmentation rather than simply being presented with a diagnosis and referral suggestion.</p><p>This segmentation technology also has enormous potential in <em>clinical training </em>as it can help professionals to learn to read medical images.</p><p>Furthermore, it can be used to <em>quantify and measure retinal pathologies</em>. Currently, retinal experts can only eyeball the differences between present and past OCT scans to objectify disease progression (eg: more intra-retinal fluids). With the AI’s automated segmentation, however, quantitative information such as the location and volume of seen anomalies can be automatically derived. This data can then be used for disease tracking and research, as an endpoint in clinical trials for example.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/661/1*QbgD1Mkw6l0NRsGMi32yRA.png" /><figcaption>left: a raw OCT scan; middle: manual segmentation; right; automated segmentation</figcaption></figure><h4>2. The Classification Network</h4><p>This second neural network analyses the tissue-segmentation maps and outputs both a diagnosis and a referral suggestion. It was trained using 14884 OCT scan volumes from 7621 patients. Segmentation maps were automatically generated for all scans. Clinical labels were obtained by examining the patient’s clinical records in order to determine retrospectively 1) the final diagnosis (after all investigations), 2) the optimal referral pathway (in light of that diagnosis).</p><p>The classification network, therefore, takes segmentation maps and learns to prioritize patients’ need for treatment into urgent, semi-urgent, routine and observation-only. It then outputs a diagnosis in form of a probability of multiple, concomitant retinal pathologies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/786/1*ovHx5v6JVrumeAigR4R3Sw.png" /><figcaption>Output: predicted diagnosis probabilities and referral suggestions</figcaption></figure><h3>Image Ambiguity and Ensembling</h3><p>Image interpretation and segmentation can be difficult for humans and machines alike due to the presence of ambiguous regions, where the true tissue type cannot be deduced from the image, and thus multiple equally plausible interpretations exist. To overcome this challenge, DeepMind’s framework uses an ensemble of 5 segmentation instances instead of 1. Each network instance creates a full segmentation map for the given scan, resulting in 5 different hypotheses. These different maps, just like different clinical experts, agree in areas with clear image structures but may differ in ambiguous low-quality regions. Using this ensemble, the ambiguities arising from the raw OCT scans are presented to the subsequent decision (classification) network. The classification network also has an ensemble of 5 instances which are applied to each of the 5 segmentation maps, resulting in a total of 25 classification outputs for every scan.</p><h3>Results:</h3><p>The framework achieved an area under the ROC curve that was over 99% for most of the pathologies, on par with clinical experts. As for the referral suggestion, its performance matched that of the five best specialists and outperformed that of the other three.</p><h3>Future:</h3><blockquote>OCT is now one of the most common imaging procedures with 5.35 million OCT scans performed in the US Medicare population in 2014 alone.</blockquote><p>The widespread availability of OCT has not been matched by the availability of expert humans to interpret scans and refer patients to the appropriate clinical care.</p><p>DeepMind’s AI solution has the potential to lower the cost and increase the availability of screening for retinal pathologies using OCT. Not only can it automatically detect the features of eye diseases, but it also prioritizes patients most in need of urgent care by recommending whether they should be referred for treatment. This instant triaging process should drastically cut down the delay between the scan and treatment, allowing patients with serious diseases to obtain sight-saving treatments in time.</p><blockquote>“Anytime you talk about machine learning in medicine, the knee-jerk reaction is to worry that doctors are being replaced. But this is not going to replace doctors. In fact it’s going to increase the flow of patients with real disease who need real treatments,”<em> Dr. Ehsan Rahimy, MD, a Google Brain consultant and vitreoretinal subspecial­ist in practice at the Palo Alto Medical Foundation.</em></blockquote><p>*Addendum April 11, 2021: this article has been renamed from its original title <em>“Google DeepMind might have just solved the “Black Box” problem in medical AI.”</em> Like some readers pointed out, the segmentation algorithm remains a black box itself and doesn’t truly solve the problem. However, from a clinical point of view, the segmented views of the 3D OCT scans are extremely valuable. It is very similar to how physicians make diagnoses from imaging: 1) identify anomalies on the scan 2) associate anomalies with a pathology. Therefore, adding an <em>intermediate representation</em> to the diagnostic algorithm significantly improves the AI system’s interpretability, making it more feasible for clinical deployment in the future.</p><h3>Read more from Health.AI:</h3><p><a href="https://medium.com/health-ai/ai-2-0-in-ophthalmology-googles-second-publication-c3b5390c19ae">AI 2.0 in Ophthalmology — Google’s Second Publication</a></p><p><a href="https://medium.com/health-ai/deep-learning-in-ophthalmology-using-128-175-retinal-images-59814e8a3f68">Deep Learning in Ophthalmology — How Google Did It</a></p><p><a href="https://medium.com/health-ai/machine-learning-and-oct-images-the-future-of-ophthalmology-47dc64ee9dc6">Machine Learning and OCT Images — the Future of Ophthalmology</a></p><p><a href="https://medium.com/health-ai/machine-learning-and-plastic-surgery-95a8d75a631d">Machine Learning and Plastic Surgery</a></p><p><a href="https://medium.com/health-ai/ai-neural-network-for-pediatric-cataract-6a4341f31263">AI &amp; Neural Network for Pediatric Cataract</a></p><p><a href="https://medium.com/health-ai/the-cutting-edge-the-future-of-surgery-9dfa4315ae4f">The Cutting Edge: The Future of Surgery</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3ed8bc21f636" width="1" height="1" alt=""><hr><p><a href="https://medium.com/health-ai/google-deepmind-might-have-just-solved-the-black-box-problem-in-medical-ai-3ed8bc21f636">Google DeepMind -Improving the interpretability of “Black Box” in medical imaging</a> was originally published in <a href="https://medium.com/health-ai">Health.AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI in Medicine — Majority decision isn’t always right]]></title>
            <link>https://medium.com/health-ai/ai-2-0-in-ophthalmology-googles-second-publication-c3b5390c19ae?source=rss-fdbfaaae6aa8------2</link>
            <guid isPermaLink="false">https://medium.com/p/c3b5390c19ae</guid>
            <category><![CDATA[health-technology]]></category>
            <category><![CDATA[ophthalmology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[healthcare]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Susan Ruyu Qi]]></dc:creator>
            <pubDate>Tue, 17 Jul 2018 14:59:27 GMT</pubDate>
            <atom:updated>2019-12-06T09:57:28.247Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qA8FiQaBZjCS2zL8.png" /></figure><p>Following their major publication in JAMA that marked a major breakthrough in both the AI and healthcare communities, Google made some fine tuning to their Deep Learning model and published their new results on <a href="https://arxiv.org/abs/1710.01711">Arxiv</a>. *This article was then published in <a href="https://www.aaojournal.org/article/S0161-6420(17)32698-2/fulltext"><em>Ophthalmology</em></a>, one of the most important journals in the field, with an impact factor of 6.1.</p><p>I had explained in a<a href="https://medium.com/health-ai/deep-learning-in-ophthalmology-using-128-175-retinal-images-59814e8a3f68"> previous article</a> how Google designed their initial AI model to detect diabetic retinopathy (DR) from fundus photos. Here’s what’s new in Google’s AI 2.0:</p><h3>1. Redefining the Gold Standard Using “Adjudication”</h3><p>Garbage in garbage out. We all know that AI is biased if trained using inaccurate labels; it can even be dangerous in medicine.</p><p>While having high-quality ground truth label is critical for training machine learning models, it is easier said than done since medicine is often subjective. Take diabetic retinopathy for example, doctors will most often agree when lesions are obvious: this one has retinopathy and that one does not. However, when asked to grade the disease on a scale of 1–5, disagreements occur. On the image below, Google showed that for the same image of the retina, different ophthalmologists will grade the image differently, consistent only around 60% of the time with themselves and with others.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*iNxSDp8c5HlUr5mZ" /><figcaption>Each row is an image, each column is an ophthalmologist grader. Colours represent the severity grades given by each ophthalmologist.</figcaption></figure><p>That is, in the most part, due to the subjective variance in exact definition of grades and boundaries between the 5 different grades. For example, while mild DR is defined as “<em>having microaneurysms</em>”, image artefacts often resemble microaneurysms and was a common source of disagreement. Moderate DR is defined as “<em>more than microaneurysms but less than severe NPDR”</em>, which is also open to interpretation. Luckily, clinical care accounts for much more than a simple image. Treatment plan is personalized for each patient according to his/her age, medical and family history, disease progression, diabetes control, and a much more thorough eye exam using other imaging modalities (OCT) and direct stereoscopic examination of the retina after dilation. Although not affecting patient care, disagreements in grading scores do make research and creating image label much more difficult.</p><h4>When doctors disagree, who is right? Who’s answer should we label as ground-truth?</h4><p>Traditionally, taking the “majority decision” has been a popular method for defining the reference standard. For example, Google had hired enough ophthalmologists to have each image of their initial dataset (128k images) read by 7–8 different people, independently. They then took the majority decision as the final label for each image. This method is flawed as it does introduce a bias where the algorithm will miss subtle findings that the majority of ophthalmologists might not identity. In their second study, Google suggested a more rigorous way to define the “gold-standard”, through adjudication. <strong>Using this method, instead of taking the majority grading when there is disagreement, doctors will decide together, face to face, to conclude on a final decision.</strong></p><p>They tested this adjudication process on a subset of around 6000 images. The images were first evaluated independently by 3 fellowship-trained retina specialists who then discussed face-to-face to resolve disagreements and determined a “final diagnosis”.</p><p>3737 images with adjudicated ground-truth labels were used as the tune set; ie: it was used for tuning the algorithm hyperparameters (e.g. image resolution, learning rate) and making model choices (e.g. network architectures), but not for training the model parameters.The rest of the images with adjudicated grading were used as the validation set.</p><p>In this study, they demonstrated that<strong> model performance was significantly improved even if only a small subset (0.22%) of the training image grades were adjudicated.</strong></p><h3>2. Upgrade from Binary Prediction to a 5-class Rating Prediction</h3><p>Instead of a binary “referable” vs “non-referable diabetic retinopathy”, in this second study, the Google team trained a 5-class prediction model that can grade an image’s disease severity: <em>none, mild, moderate, severe, and proliferative</em>. This is in accordance to the most commonly used International Clinical Diabetic Retinopathy (ICDR) disease severity scale. This makes the model more suitable for clinical practice.</p><h3>3. Bigger Data Set (1.6M Images)</h3><p>The training portion of the development set was increased from 128 175 images in their first study to over 1.6M images in this study. It contains fundus images from 238 610 patients.</p><p><strong>Input resolution </strong>in this new model is<strong> </strong>779 x 779 pixels, a large increase over the 299 x 299 pixels used in their previous study. The <strong>model architecture</strong> was also upgraded, from Inception-v4 to Inception v316.</p><p>Overall, for each of the five gradings, the algorithm has <strong>AUC values between 0.986 to 0.998. </strong>In other words, the algorithm is capable of labelling all five grades of the disease with high sensitivity and specificity!</p><h3><strong>Future: an Increased Need for Ophthalmologists:</strong></h3><p>When tele-ophthalmology was first introduced, it allowed patients to be screened remotely using fundus photos, making care much accessible to rural populations. Consequently, it has also allowed more diabetic retinopathy cases to be diagnosed and referred to retina specialists for closer examination, treatment and followups. This has overwhelmed many ophthalmology clinics’ already busy schedule and wait-lists. The AI algorithm, through high productivity and low cost, will make screening even more accessible to patients worldwide.</p><blockquote>Diabetic retinopathy is the leading cause of blindness in working-age adults.</blockquote><p>DR is an insidious disease that slowly damages the retina, leading to symptoms only in late-stage, when damages have become irreversible. Current guidelines therefore recommend yearly screening to all diabetic patients. In real life however, <strong>less than one third of patients are getting this required exam </strong>(Ontario, Canada).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*3ktqkyq8mUYfAFS7" /></figure><blockquote>By 2040, it is approximated that 600 million people will have diabetes, with one-third expected to have diabetic retinopathy. -Ting et al. JAMA.</blockquote><p>With 600 million people all requiring yearly screening, along with an increased life expectancy, and an aging population, the demand for ophthalmologists will be higher than ever. As a future ophthalmologist, I’m glad that AI algorithms now exist to assist me in the future, freeing me from the repetitive pattern-recognition tasks, and can allow me to focus on patient care and innovative research.</p><h3>Read more from Health.AI:</h3><p><a href="https://medium.com/health-ai/deep-learning-in-ophthalmology-using-128-175-retinal-images-59814e8a3f68">Deep Learning in Ophthalmology — How Google Did It</a></p><p><a href="https://medium.com/health-ai/machine-learning-and-oct-images-the-future-of-ophthalmology-47dc64ee9dc6">Machine Learning and OCT Images — the Future of Ophthalmology</a></p><p><a href="https://medium.com/health-ai/machine-learning-and-plastic-surgery-95a8d75a631d">Machine Learning and Plastic Surgery</a></p><p><a href="https://medium.com/health-ai/ai-neural-network-for-pediatric-cataract-6a4341f31263">AI &amp; Neural Network for Pediatric Cataract</a></p><p><a href="https://medium.com/health-ai/the-cutting-edge-the-future-of-surgery-9dfa4315ae4f">The Cutting Edge: The Future of Surgery</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c3b5390c19ae" width="1" height="1" alt=""><hr><p><a href="https://medium.com/health-ai/ai-2-0-in-ophthalmology-googles-second-publication-c3b5390c19ae">AI in Medicine — Majority decision isn’t always right</a> was originally published in <a href="https://medium.com/health-ai">Health.AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Machine Learning and OCT Images — the Future of Ophthalmology]]></title>
            <link>https://medium.com/health-ai/machine-learning-and-oct-images-the-future-of-ophthalmology-47dc64ee9dc6?source=rss-fdbfaaae6aa8------2</link>
            <guid isPermaLink="false">https://medium.com/p/47dc64ee9dc6</guid>
            <category><![CDATA[medtech]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[ophthalmology]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[health]]></category>
            <dc:creator><![CDATA[Susan Ruyu Qi]]></dc:creator>
            <pubDate>Fri, 08 Dec 2017 17:59:58 GMT</pubDate>
            <atom:updated>2018-08-23T04:59:38.577Z</atom:updated>
            <content:encoded><![CDATA[<p><em>“Optical coherence tomography (OCT) has become the most commonly used imaging modality in ophthalmology, with 4.39 million, 4.93 million, and 5.35 million OCTs performed in 2012, 2013, and 2014, respectively, in the U.S. Medicare population.” — </em><a href="https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Medicare-Provider-Charge-Data/Physician-and-Other-Supplier.html"><em>Cms.gouv</em></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/279/1*viHadRmW2dey0T52Xp6WYg.gif" /><figcaption>OCT scan of the retina</figcaption></figure><h3>1) AI for the diagnosis of AMD</h3><p>Age-Related Macular Degeneration (AMD) is a common eye condition and a leading cause of vision loss among people age 50 and older. It causes damage to the macula, a part of the retina, causing loss of central vision. It affects all “straight-ahead” activities, such as reading, driving, seeing faces, etc. Imagine looking at a clock with hands, patients with AMD might see the clock’s numbers but not the hands (Figure 1).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/180/0*yoyWrsPTwHMJ5eDF." /><figcaption>A clock seen by AMD patients. Central vision is affected but peripheral vision is spared.</figcaption></figure><p>There are two major types of AMD: dry and wet. <strong>Dry AMD</strong> accounts for 80% of patients. It’s the less urgent of the two as it is slowly progressive. On the other hand, <strong>Wet AMD</strong> is much more serious. It has a much faster progression and can lead to vision loss if treatment is not promptly initiated. The underlying disease in Wet AMD is the formation of abnormal blood vessels on the retina. These vessels can leak blood and cause inflammation. This will cause scarring of the macula (retina) and subsequent vision loss.</p><p>Fortunately, we’ve found treatments that work quite well in Wet AMD. These are intravitreal injections — injections inside the posterior chamber of the eye — of agents that stop the growth and formation of bad blood vessels. They are also commonly called anti-VEGF Injections (anti-Vascular Endothelial Growth Factor).</p><p>Correct classification of the type of AMD is therefore crucial. We want to initiate injections promptly in those with Wet AMD.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*LLCZxXfuHS5hmYkS." /><figcaption>Left: Wet AMD with fluid buildup at the macula | Right: post-injection, improved macula, less fluid</figcaption></figure><p>This problem turns out to be a great fit for computer vision and machine learning. Researchers from University of Washington trained a modified version of the VGG16 convolutional neural network (CNN) using 2.6 million OCT images of 43,328 macular OCT scans from 9285 patients. Their CNN is able to correctly distinguish AMD from normal OCT images with an accuracy of 93%. Peak sensitivity and specificity with optimal cutoffs were 92.64% and 93.69%.</p><p>Read more here:<a href="http://www.sciencedirect.com/science/article/pii/S2468653016301749"> Deep Learning Is Effective for Classifying Normal versus Age-Related Macular Degeneration OCT Images</a></p><h3>2) AI &amp; Anti-VEGF Injections in AMD</h3><p>After correctly identifying AMD in the retina, the physicians were challenged with another question: when is the best time to give anti-VEGF injections? OCT scans again play a major role in this treatment decision.</p><p>Ideally, we want to use the least amount of injections while maintaining no active blood vessel leakage in the retina. It turns out machine learning can also be helpful in addressing this problem.</p><p>In this study, the authors used 183,402 retinal OCT B-scans to train a GoogLeNet inception deep convolutional neural network to solve the binary classification problem of whether a patient needs an injection or not. Two groups were created: 1) an “injection group” containing OCT images that have a following intravitreal injection during the first 21 days after image acquisition. 2) the rest without an injection is labelled as “no injection”. After training, the CNN had a prediction accuracy of 95.5%. For single retinal B-scans in the validation dataset, a sensitivity of 90.1% and a specificity of 96.2% were achieved.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/554/1*J9sfbE6iX0QEBa6WnQ0l_Q.png" /><figcaption>a) Injection group | b) No-injection group</figcaption></figure><p>Read more here:<a href="https://link.springer.com/article/10.1007/s00417-017-3839-y?no-access=true"> OCT-based deep learning algorithm for the evaluation of treatment indication with anti-vascular endothelial growth factor medications</a></p><h3>AI applications in ophthalmology is still in its infancy.</h3><p>The above shows applications of automated image classification using historical clinical data. Thus, the algorithms are learning from historical physician decisions, including both the good and less good ones. However, we can imagine that much more can be done using this technology in the future. For instance, we can feed the algorithm with outcome data — such as improvement or deterioration of visual acuity — so that it can make even more intelligence predictions to inform physicians of potential treatment outcomes.</p><p>While the accuracy of these models are incredibly impressive, we need to remain prudent and sober when considering how to deploy these systems to the real world, especially as it applies to critical applications in healthcare. Today’s limitations of machine learning include a poor understanding of what the model is <em>actually</em> doing. As far as we know, <em>the classifier output is not the result of a reasoning that leads to a decision, but rather a computation based on prior clinical treatment data. </em>It can provide valuable information to clinicians, acting like a second opinion to trained physicians. However, care should be taken to avoid acting on results without final human approval.</p><p>Read more: <a href="https://medium.com/health-ai/deep-learning-in-ophthalmology-using-128-175-retinal-images-59814e8a3f68">Deep Learning in Ophthalmology — How DeepMind Did It</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=47dc64ee9dc6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/health-ai/machine-learning-and-oct-images-the-future-of-ophthalmology-47dc64ee9dc6">Machine Learning and OCT Images — the Future of Ophthalmology</a> was originally published in <a href="https://medium.com/health-ai">Health.AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deep Learning in Ophthalmology —How Google Did It]]></title>
            <link>https://medium.com/health-ai/deep-learning-in-ophthalmology-using-128-175-retinal-images-59814e8a3f68?source=rss-fdbfaaae6aa8------2</link>
            <guid isPermaLink="false">https://medium.com/p/59814e8a3f68</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[medicine]]></category>
            <category><![CDATA[ophthalmology]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Susan Ruyu Qi]]></dc:creator>
            <pubDate>Tue, 26 Sep 2017 22:06:30 GMT</pubDate>
            <atom:updated>2018-08-23T07:15:57.997Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>Ophthalmology is arguably the best “test bed” for AI techniques within the healthcare space. The sheer volume of high quality data due and advanced imaging methods makes ophthalmology a clear candidate for the advances in computer vision.</blockquote><p>A recent project by Google Research is a perfect example: They used over 120,000 retinal images to train a neural network to detect diabetic retinopathy, a leading cause of blindness. <strong>The resulting model can match the performance of ophthalmologists</strong>. This study, published in JAMA earlier this year, was a major breakthrough in both the AI and the health-tech world. Let’s look at the project in more detail.</p><blockquote><strong>Objective: To use deep learning for the automated detection of diabetic retinopathy (DR) and diabetic macular edema (DME) in retinal fundus photographs.</strong></blockquote><p><strong>What is diabetic retinopathy (DR)?</strong> It’s the fastest growing cause of blindness, affecting more than 20% of the 488 millions of people living with diabetes worldwide. High blood sugar can cause damage to blood vessels of the retina (tissue covering the back of the eye, made up of light-sensitive cells.) Vision is not affected initially but irreversible blindness will occur in time without treatment. Early detection is therefore crucial in order to administer timely treatment and prevent the disease’s progression.</p><blockquote><strong>Diabetic Retinopathy leads to irreversible blindness. If caught early, it can be treated. However, a patient may experience no symptoms early on, making regular screening vital.</strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/726/1*jsJs4sKOywNqNuQ-rijKKw.png" /><figcaption>Retina is the tissue at the back of the eye. It’s made of cells that capture light signals and communicate them to the brain.</figcaption></figure><p><strong>Early detection, how? </strong>Diagnosis of DR requires direct visualization of the retina by medical specialists, either via eye exam or imaging. Specialists then grade the level of disease based on presence of characteristic lesions, indicative of blood vessel damage (hemorrhage, microaneurysms, exudates, ie: bleeding or fluid leakage). However, eye specialists are not available in many parts of the world where diabetes is prevalent. Can machines help?</p><p><strong>Machine Learning:</strong> Indeed, machine learning had been used in the past, but unlike in this Google project, it was mostly focused on <em>“feature engineering,”</em> which involves computing explicit features specified by experts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/926/1*d03nrREZ0uIMo5mo_mNn0w.png" /><figcaption>Machine Learning “Feature engineering”— detecting pre-specified features</figcaption></figure><p><strong>What is Deep Learning and how can it help?</strong> Deep Learning is a machine learning technique that uses neural networks to model very complex functions. Luckily, there is a well-developed algorithm called <em>“back-propagation”</em> that allows the neural network to learn its parameters automatically given training data. Typically, the more data we feed into the training algorithm, the better the neural net performs.</p><p>In other words, instead of programming the machine to detect “1,2,3” (signs) in order to make diagnoses “A or B” (diseased or normal), we now tell machines that these images are A (diseased), the others are B (normal); learn from them and figure out why.</p><h3><strong>How did Google do it?</strong></h3><ol><li>They used <em>convolutional neural network</em>. It’s a specific type of neural network optimized for image classification, the same technique that Google uses to label millions of Web images. In other words it’s a large mathematical function with millions of parameters. Deep learning is simply the process of training this function.</li><li><strong>Training Data Set: 128 175</strong> retinal images from EyePACS (electronic medical record) in the US and 3 eye hospitals in India. All images were graded by 3 to 7 different ophthalmologists, from a panel of 54 US-licensed ophthalmologists and senior residents. Gradings included DR, DME and image quality. These serve as labels to the images.</li><li><strong>Validation Sets x2: </strong>After this training, the algorithm is put to the test with 2 validation sets: 1) A random sample of <strong>9963</strong> images taken from EyePACS, not overlapping with the previous training set. 2) A publicly available data set called Messidor-2, with <strong>1748</strong> images.</li><li>The machine’s grading using the algorithm is then compared to that of the ophthalmologists’.</li></ol><h4><strong>How did the Deep Learning algorithm do? — Extremely well.</strong></h4><p>In the setting of disease screening, we want to aim for high sensitivity (allowing us to confidently rule out the negatives; i.e. very little false negative). In that setting, the algorithm achieved <strong>96–97% sensitivity and 93% specificity</strong>! That is comparable to, if not slightly better than, the results of the 8 ophthalmologists grading the images. Since these 8 ophthalmologists were already elites of the initial group of 54 specialists (they were selected based on high rate of self-consistency), the model is doing extremely well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/553/1*F2n8qNCSIDqWYdQ7vjwILw.png" /><figcaption>Validation Set 1 Performance (black line = algorithm; colored dots = ophthalmologists)</figcaption></figure><blockquote>Take home message: Artificial Intelligence will change the way we practice medicine. This deep learning algorithm is already doing better at diagnosing diabetic retinopathy than many physicians. Yet, it is only the first step in bringing AI to ophthalmology. This is the future. Stay tuned!</blockquote><ul><li>* This research, although revolutionary, is not yet mature enough to be implemented in the clinical setting. That is mainly because the algorithm currently only detects DR and not the other common eye pathologies. Its limitations will be discussed in the a separate article.</li></ul><p>Google has since updated their algorithm. See what’s new <a href="https://medium.com/me/stories/public">here</a>.</p><p>Read more here: <a href="https://medium.com/health-ai/machine-learning-and-oct-images-the-future-of-ophthalmology-47dc64ee9dc6">Machine Learning and OCT Images- the Future of Ophthalmology</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=59814e8a3f68" width="1" height="1" alt=""><hr><p><a href="https://medium.com/health-ai/deep-learning-in-ophthalmology-using-128-175-retinal-images-59814e8a3f68">Deep Learning in Ophthalmology —How Google Did It</a> was originally published in <a href="https://medium.com/health-ai">Health.AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI & Neural Network for Pediatric Cataract]]></title>
            <link>https://medium.com/health-ai/ai-neural-network-for-pediatric-cataract-6a4341f31263?source=rss-fdbfaaae6aa8------2</link>
            <guid isPermaLink="false">https://medium.com/p/6a4341f31263</guid>
            <category><![CDATA[big-data]]></category>
            <category><![CDATA[ophthalmology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[health]]></category>
            <dc:creator><![CDATA[Susan Ruyu Qi]]></dc:creator>
            <pubDate>Tue, 25 Apr 2017 03:25:13 GMT</pubDate>
            <atom:updated>2017-09-03T22:59:49.080Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>Researchers from China have developed an artificial intelligence (AI) platform that can diagnose congenital cataracts as accurately as ophthalmologists.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QqJiB2iWY_xk6MqVSkFvsg.png" /></figure><h3><strong>What is Cataract?</strong></h3><p>A cataract is clouding of the natural lens of the eye. It is extremely common among the elderly population but can also affect infants at birth or develop at a young age.</p><blockquote>“In children, cataract causes more visual disability than any other form of treatable blindness. Children with untreated, visually significant cataracts face a lifetime of blindness at tremendous quality of life and socioeconomic costs to the child, the family, and the society”- American Academy of Ophthalmology (AAO).</blockquote><h3><strong>Detecting Cataracts in Children:</strong></h3><p>The asymptomatic progression of pediatric cataracts makes it difficult for early detection. Not to mention that young kids can’t really express themselves. However, catching the disease early is critical; late diagnosis often result in poor visual outcomes.</p><h3><strong>Problems:</strong></h3><ul><li>Diagnosis requires expertise, especially in children and neonates.</li><li>There is a relatively small number of ophthalmologists for the size of our population; with an even smaller number of pediatric ophthalmologists.</li><li>Experts are concentrated in specialized centers but their coverage is largely insufficient.</li><li>Rural areas get almost no coverage.</li><li>This lack of accessibility causes lifetime vision loss to affected children.</li></ul><h3><strong>Solution: AI</strong></h3><p>A team in China tried to address the problem of lack of pediatric ophthalmologists by using Artificial Intelligence (AI), specifically deep learning and convolutional neural networks. They built an AI machine, CC-Cruiser, that has three main functions: 1) Diagnosis 2) Risk stratification 3) Treatment suggestions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/582/1*DQEubWGPU8WFdMvwO_J5Og.png" /><figcaption>Screening for Congenital Cataract using Artificial Intelligence</figcaption></figure><p>Their initial work was done on congenital cataract cases only (amongst other types of pediatric cataracts) but further studies are ongoing.</p><h3>Future: AI requires Big Data</h3><p>A cloud-based platform has been created for multi-hospital collaboration; Doctors at hospitals around the country can access the platform and upload patient images. More data will be collected this way, which is essential to AI’s learning and fine-tuning.</p><blockquote>“The limited resources of patients and the isolation of the data in individual hospitals represent a bottleneck in data usage. Building a collaborative cloud platform for data integration and patient screening is an essential step.”-Professor Lin Haotian of Sun Yat-Sen University</blockquote><p>Liked what your read? Don’t forget to click like below!<br>Read more about the <a href="http://healthainow.com/ai-neural-network-for-pediatric-cataract/">future of medicine</a> at <a href="https://medium.com/health-ai">Health.AI</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6a4341f31263" width="1" height="1" alt=""><hr><p><a href="https://medium.com/health-ai/ai-neural-network-for-pediatric-cataract-6a4341f31263">AI &amp; Neural Network for Pediatric Cataract</a> was originally published in <a href="https://medium.com/health-ai">Health.AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Diagnosing Dry Eyes with Computer Eyes]]></title>
            <link>https://medium.com/health-ai/diagnosing-dry-eyes-using-machine-learning-infrared-thermal-images-1b344fe988eb?source=rss-fdbfaaae6aa8------2</link>
            <guid isPermaLink="false">https://medium.com/p/1b344fe988eb</guid>
            <category><![CDATA[health]]></category>
            <category><![CDATA[ophthalmology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[medicine]]></category>
            <dc:creator><![CDATA[Susan Ruyu Qi]]></dc:creator>
            <pubDate>Sat, 15 Apr 2017 03:52:19 GMT</pubDate>
            <atom:updated>2017-09-03T22:57:39.991Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>An application of AI &amp; I</em><strong><em>nfrared Thermal Images in Ophthalmology</em></strong></h4><p>Have you ever had that itching or stinging sensation in your eyes, after staring at a computer screen for too long or when wearing contact lenses all day? Congratulations, you might have a condition called “dry eye syndrome”.</p><p>Dry Eye disease is a common condition affecting 8% of the population caused by decreased quantity or quality of tears. If left untreated, this condition can lead to pain, ulcers, even scars on the cornea. Therefore, rapid diagnosis is important and is made clinically by 1)Measurement of tear production-(ie: <em>quantity),</em> 2)Evaluation of tear film stability -(ie: <em>quality).</em></p><p>Unfortunately, both of these tests are quite unpleasant:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/257/1*K_5XB4YKS_-LkME08_7hvw.png" /><figcaption>Schirmer’s test</figcaption></figure><blockquote><em>1) </em><strong><em>Schirmer’s test </em></strong><em>— </em>measure of tear production. [filter paper is placed inside the lower lid of eye to quantify the amount of wetting of the strip over a 5 min duration.] Yes, 5 minutes…</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/209/1*YAnpC9PakTyfYtwrZhAMCw.png" /><figcaption>Fluorescein dye applied to the eye</figcaption></figure><blockquote><em>2)</em><strong><em>Tear breakup time</em></strong><em> (TBUT) — </em>evaluation of tear film stability. [fluorescein drops are applied onto the ocular surface; the time needed for dry spots to appear on the cornea is assessed using a green fluorescent light.]</blockquote><p>Both of these tests are subjective, invasive, and uncomfortable.</p><p>In a recent study, researchers used <strong>infrared thermal images</strong> of the eye along with <strong>machine learning algorithms</strong> Gabor transform and Discrete Wavelet Transform (DWT)* to detect Dry Eyes. These machine learning methodologies are used to extract features from specific frames of the infrared thermography images, which is further segmented into eye regions and the data is analyzed accordingly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*NnjfS62xQJc-tfoTsz1Cyg.png" /></figure><p>This method offers several advantages:</p><ol><li>It is semi-automatic, therefore less susceptible to inter-observer variability.</li><li>It is more accurate than standard clinical tools.</li><li>It does not require special dye, and is more convenient for the patient.</li></ol><ul><li><em>Gabor transform and Discrete Wavelet Transform are methodologies for automatic feature extraction from biomedical images.</em></li></ul><p>Read More Here: <a href="http://healthainow.com/diagnosing-dry-eyes-with-computer-eyes/">Evaluation of Evaporative Dry Eye Disease Using Thermal Images of Ocular Surface Regions with DWT and Gabor Transform</a></p><p>Next up in Health.AI: <a href="https://medium.com/health-ai/heat-screening-for-breast-cancer-computer-vision-for-dieting-brain-wave-passwords-c5294b5e73b">Heat Screening for Breast Cancer, Computer Vision for Dieting, Brain Wave Passwords</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1b344fe988eb" width="1" height="1" alt=""><hr><p><a href="https://medium.com/health-ai/diagnosing-dry-eyes-using-machine-learning-infrared-thermal-images-1b344fe988eb">Diagnosing Dry Eyes with Computer Eyes</a> was originally published in <a href="https://medium.com/health-ai">Health.AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>