<?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 Dr. Valerio Rizzo on Medium]]></title>
        <description><![CDATA[Stories by Dr. Valerio Rizzo on Medium]]></description>
        <link>https://medium.com/@VRizzo?source=rss-158e5d9662f1------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*DxltGzMvQEaQuyRkcwTH_g.jpeg</url>
            <title>Stories by Dr. Valerio Rizzo on Medium</title>
            <link>https://medium.com/@VRizzo?source=rss-158e5d9662f1------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 12:15:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@VRizzo/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[Non Omnis Moriar: How AI will help us in the Quest to Restore Memories]]></title>
            <link>https://medium.com/@VRizzo/non-omnis-moriar-5002e16dc95b?source=rss-158e5d9662f1------2</link>
            <guid isPermaLink="false">https://medium.com/p/5002e16dc95b</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[human-physiology]]></category>
            <category><![CDATA[neuroscience]]></category>
            <dc:creator><![CDATA[Dr. Valerio Rizzo]]></dc:creator>
            <pubDate>Mon, 18 Sep 2023 12:01:47 GMT</pubDate>
            <atom:updated>2023-09-18T16:21:07.288Z</atom:updated>
            <content:encoded><![CDATA[<blockquote><strong>“Èxegì</strong> <strong>monumèntum</strong> àere perènnius</blockquote><blockquote>règalìque sitù pýramidum àltius,</blockquote><blockquote>quòd non ìmber edàx , nòn Aquilo ìmpotens</blockquote><blockquote>pòssit dìruere àut ìnnumeràbilis</blockquote><blockquote>ànnorùm seriès èt fuga tèmporum”</blockquote><p><strong>Orazio, “Odi” (III, 30), “Exegi monumentum”</strong></p><p>The Latin phrase “Non Omnis Moriar,” taken from Horace’s “Odes” (III, 30), resonates deeply with me. In this ode, Horace declares that he will never wholly die but will continue to live on, fresh and young, through the praise he receives from future generations. This belief, that a part of us remains eternal despite adversities, is a sentiment I’ve held close to my heart, especially when thinking of my grandmother. After a devastating brain stroke, my grandmother’s cognitive faculties were severely affected. Her memory of both distant and recent events became fragmented, and her once-vibrant personality was altered. Amidst her pronounced echolalia, a condition where she would often repeat nonsensical phrases, I once asked her in desperation, “Nonna, what’s going on in your mind?” Her response, “Nothing is going on in my mind…There is nothing in my mind,” was a haunting reminder of the void the stroke had created. Yet, I’ve always clung to the idea that response was a glimpse, an intact leftover of who she was, somewhere deep within trapped in a tangle of cognitive malfunctions, unable to put back together neatly all those pieces of autobiographical and emotional memories defining who we are.</p><p>This personal experience has always ignited my passion to delve into the intricacies of memory and the promising advancements in neuroscience and AI.</p><p><strong>The Lullaby of Memory</strong></p><p>In 2015, Jakobsen and colleagues published a compelling article titled “Why musical memory can be Preserved in advanced Alzheimer’s Disease”. The research delved into the fascinating resilience of musical memory in Alzheimer’s patients. In this study, researchers used 7T fMRI to capture those detailed neural activity patterns that are less visible in lower-resolution scans. This level of detail was crucial when trying to understand the nuanced neural pathways involved in musical memory. Historical studies, such as those by Wilder Penfield, have pointed towards the temporal cortex’s role in encoding episodic musical memory, but in this new study, researchers have found that regions like the caudal anterior cingulate and the ventral pre-supplementary motor area play a pivotal role in long-term musical memory and surprisingly relatively well preserved in patients with Alzheimer’s disease. Therefore, some of our memories might survive even critical neurobiological deteriorations, but how to bring them back? Is there a way to capture them?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ssk0NwUI-xxWctxdIM_x9A.png" /><figcaption><em>The </em>musical memory region of interest (red) on a 3D brain render<em> highlighting the caudal anterior cingulate and the ventral pre-supplementary motor area. </em>Image Credits to the authors of the original paper (<a href="https://academic.oup.com/brain/article/138/8/2438/330016">source</a>)</figcaption></figure><p><strong>The AI Symphony: Decoding the Neural Notes</strong></p><p>The recent study on AI’s application to electroencephalography (EEG) music reconstruction by Bellier and colleagues is a testament to the convergence of neuroscience and artificial intelligence. As someone deeply invested in both fields, this research resonated with me, offering a glimpse into the future of memory preservation and restoration.</p><p><strong>Brain Circuitry and AI’s Role</strong></p><p>The study’s primary focus was the superior temporal gyri (STG), crucial for processing auditory information, and translating the sounds we hear into comprehensible units. The STG’s role in music perception is particularly significant in the context of memory, as it has been implicated in encoding episodic musical memories. The anteroposterior parcellation in the STG is particularly significant as it differentiates between various types of auditory stimuli, allowing for nuanced processing.</p><p>The researchers employed a unique AI algorithm to decode and reconstruct music pieces that participants listened to, directly from their EEG recordings. This approach is groundbreaking, as it moves beyond mere recognition to actual reconstruction.</p><p><strong>Methodology and Paradigm</strong></p><p>Participants listened to a specific rock song (“Another Brick in the Wall, Part 1” by Pink Floyd): why the researchers chose this song and how the participants of the experiments liked that song is unknown, but I like to think they all had <strong>very </strong>good musical taste! However, subjets’ neural responses were recorded using EEG, or better, the authors actually used “2,668 ECoG electrodes implanted in 29 neurosurgical patients”. This is different from EEG as in this case electrodes are directly placed on the surface of the brain to record electrical activity. Electrocorticography (ECoG), a type of intracranial electroencephalography (iEEG), provides a much higher spatial resolution, allowing for more precise localization of neural activity and since the electrodes are placed directly on the brain, ECoG bypasses the interference caused by the skull and scalp, leading to clearer and more robust signal recordings.</p><p>The auditory spectrogram of the song, which represents the frequency content of the song over time, was used as a reference for the AI model.</p><p><strong>Plugging the AI Amplifier</strong></p><p>Before feeding the iEEG data into the AI model, preprocessing is essential. Raw iEEG data can be noisy, and it’s crucial to filter out irrelevant signals. The data is then segmented into epochs corresponding to specific time windows of the song, ensuring that the AI model is trained on consistent and relevant data.</p><p>The study utilized Spectrotemporal Receptive Fields (STRFs) to analyze tuning patterns. Independent Component Analysis (ICA) was employed to highlight electrode populations tuned to distinct STRF features. Initially, an ICA with 10 components was run on the centered STRF coefficients to identify components individually explaining significant variance. This method helped in isolating neural populations that were sensitive to specific musical elements.</p><p>The decoding process involved comparing the actual neural responses to predicted responses based on the song’s acoustics. A bootstrapping approach was adopted, where subsets of electrodes were randomly constituted to be used as inputs for linear decoding models. The decoding process involved comparing the actual neural responses to predicted responses based on the song’s acoustics. The closer the match, the more accurate the reconstruction. The study explored both linear and nonlinear models. Linear models are based on the assumption that there’s a direct proportionality between the input (iEEG data) and the output (reconstructed music). However, the brain’s response to music might not always follow this straightforward relationship. Nonlinear models, on the other hand, can capture these intricate relationships, modeling the complexities of how the brain processes music.</p><p>To assess the statistical significance of the models, the study employed a permutation-based approach. By randomly permuting indices of the original song excerpts, a null distribution was created. The mean normalized rank of the decoding models was deemed significant if they exceeded the 97.5 percentile of this null distribution.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*M86xvfhgKyKhjwY8TOW7pA.png" /><figcaption>Protocol, data preparation, and encoding model fitting. Image Credits to the authors of the original paper (<a href="https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3002176#pbio.3002176.s009">source</a>)</figcaption></figure><p><a href="https://journals.plos.org/plosbiology/article/file?type=supplementary&amp;id=10.1371/journal.pbio.3002176.s005">Here</a> you can listen to the original song waveform transformed into a magnitude-only auditory spectrogram, then transformed back into a waveform using an iterative phase-estimation algorithm.</p><p>You can compare it with the reconstructed song excerpt obtained using <a href="https://journals.plos.org/plosbiology/article/file?type=supplementary&amp;id=10.1371/journal.pbio.3002176.s006">linear models</a> or <a href="https://journals.plos.org/plosbiology/article/file?type=supplementary&amp;id=10.1371/journal.pbio.3002176.s007">non-linear</a> models fed with all 347 significant electrodes from all 29 patients.</p><p>The result of the non-linear model reconstruction is pretty impressive, isn’t it?</p><p><strong>Conclusion</strong></p><p>The convergence of neuroscience and AI is not just about the tools but the methodologies. Data-driven approaches, combined with domain expertise in neuroscience, can lead to more accurate models (or better data) and thus insights. As we continue to refine and improve AI models and gather more data, the potential applications extend beyond memory research to areas like brain-computer interfaces, neurorehabilitation, and more. As we stand on the edge of these scientific advancements, we must wonder: How will these discoveries shape the future of memory research and treatment?</p><p>I like to imagine a future where we can tap into preserved musical memories, using AI to reconstruct and playback beloved melodies, aiding in memory recall and/ or reducing cognitive decline, or at least giving a beacon of peace and joy for those suffering souls lost in the void where the disease has confined them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/820/1*40_wpcps3oFuGGhzSCofYw.png" /><figcaption>Screen cap from Charlie Brooker’s Black Mirror (Source Netflix)</figcaption></figure><p><strong>References:</strong></p><ol><li>Jacobsen, J. H., Stelzer, J., Fritz, T. H., Chételat, G., La Joie, R., &amp; Turner, R. (2015). Why musical memory can be preserved in advanced Alzheimer’s disease. <em>Brain : a journal of neurology</em>, <em>138</em>(Pt 8), 2438–2450. <a href="https://doi.org/10.1093/brain/awv135">https://doi.org/10.1093/brain/awv135</a></li><li>PENFIELD W. (1955). The twenty-ninth Maudsley lecture: the role of the temporal cortex in certain psychical phenomena. <em>The Journal of mental science</em>, <em>101</em>(424), 451–465. <a href="https://doi.org/10.1192/bjp.101.424.451">https://doi.org/10.1192/bjp.101.424.451</a></li><li>Bellier L, Llorens A, Marciano D, Gunduz A, Schalk G, Brunner P, et al. (2023) Music can be reconstructed from human auditory cortex activity using nonlinear decoding models. PLoS Biol 21(8): e3002176. <a href="https://doi.org/10.1371/journal.pbio.3002176">https://doi.org/10.1371/journal.pbio.3002176</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5002e16dc95b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>