<?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[Brain Waves - Medium]]></title>
        <description><![CDATA[Seer makes epilepsy diagnostics and management easy by combining medical expertise with the latest technology to improve healthcare for everyone. Here our engineers and designers share technical insights. - Medium]]></description>
        <link>https://medium.com/seer-medical?source=rss----19591ed385e2---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Brain Waves - Medium</title>
            <link>https://medium.com/seer-medical?source=rss----19591ed385e2---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 04:31:12 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/seer-medical" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[A very Seer Lunar New Year]]></title>
            <link>https://medium.com/seer-medical/a-very-seer-lunar-new-year-d9c0fae84e4a?source=rss----19591ed385e2---4</link>
            <guid isPermaLink="false">https://medium.com/p/d9c0fae84e4a</guid>
            <category><![CDATA[lunar-new-year]]></category>
            <category><![CDATA[seer-medical]]></category>
            <category><![CDATA[culture]]></category>
            <dc:creator><![CDATA[Evena]]></dc:creator>
            <pubDate>Thu, 17 Mar 2022 02:27:23 GMT</pubDate>
            <atom:updated>2022-03-17T02:27:22.917Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Celebrating Lunar New Year in the digital workspace</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V77VkQnZ85haYPqQ3bIm-A.png" /><figcaption>A still of Chinese New Year lanterns</figcaption></figure><p>Seer is a pretty diverse workplace. I recall feeling nervous about feeling potentially out of place (as a lone Chinese Malaysian) when I first joined Seer as a UX Designer, but quickly noticed that my colleagues came in a range of shapes and colours, from all around the world. We have a friendly Senior Product Designer from Mexico. One of our Graphic Designers is from Indonesia. My previous manager was Sri Lankan. My Design Lead is from Israel. We have a Tech Lead from Sweden, and a software developer from Singapore (my own distinctly Malaysian drawl comes out whenever we chat). Before I knew it, I felt right at home, and sometimes I forget that I look and sound quite different to everyone else. In a strange irony, my differences were “commonplace” at Seer. My cultural identity just became part of the fabric, and wasn’t something I consciously thought about, especially in a predominantly Western country.</p><h3><strong>The birth of the “joy-luck-club”</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fbOAxCX9_vxYiJDtj_otQA.png" /><figcaption>We called ourselves the “joy-luck-club”</figcaption></figure><p>Early February this year, a Slack post was put out to acknowledge Lunar New Year, and to call for people who celebrated it. I found myself thinking, <em>“Oh that’s right, this is a big deal back home.” </em>I briefly had flashes of streets being decked out in red lanterns, and grand Zodiac animal themed installations in shopping malls, complete to blaring Lion Dances. At home, my mother would be decorating the house in her own DIY decorations — fish shapes, fans, and lanterns made out of vibrant <em>ang baos</em> (red packets). My father would play his trumpet in his study, alternating between jazz and Chinese music. Family dinners were of course, a natural part of the tradition. We would go to a Chinese restaurant, the ones with Lazy Susans, and cutlery served in a shallow dish of hot water, so you could wash them yourself first. And oh, <em>the food.</em> I particularly miss the tenderly steamed fish with ginger and garlic soy. I would try to sneakily spin the Lazy Susan so the fish head faced me, for good luck.</p><p>Under normal circumstances, I would be back home, celebrating with my family and stuffing myself silly. This would be the third consecutive year where I haven’t been able to go home. Admittedly, I had been out of touch with my roots since then, so this Slack message was a nice reminder.</p><p>I quickly joined the little emoji party of people who reacted with oranges and red packets, and soon found myself in a private channel with other people from Seer, who shared this cultural touchpoint. We had people from Vietnam, Hong Kong, Taiwan, Indonesia, China, and of course, myself, from Malaysia. The initial idea was to share some photos in a slideshow and have someone do a speech during our company wide meeting. Strangely buoyed by this sudden spike of cultural pride, I volunteered to make a video compilation of how we celebrate Lunar New Year.</p><p>Suddenly, I had 24 hours to compile clips and photos, cut and edit, find the right music, and somehow produce a fitting tribute to Lunar New Year.</p><h3>Making the video</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PsKJ8QJrBLwCUVyngL6XkA.png" /><figcaption>A still from the video featuring Bánh chưng</figcaption></figure><p>It was a frenzied night of editing, but I have no regrets.</p><p>As we collectively shared across our photos and Lunar New Year videos, I was reminded of how beautifully connected we all were in our culture. I learnt that Vietnamese Lunar New Year is called Tết, and picked up a dish I’d like to try myself, like Bánh chưng; glutinous sticky rice with mung bean and pork — which is similar to Zongzi, what I call it in Chinese. I noted that the original pronunciations for red packets in Vietnamese and Cantonese sounded strikingly similar, <em>lixi</em> and<em> lai shi. </em>At the heart of it all, Lunar New Year for us was about family, and coming together over warm meals and laughter, no matter our differences or how far apart we were.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j5mSzJ85ME5e9LKbXkABKg.png" /><figcaption>A childhood family photo shared from Sana, our Talent Acquisition Sourcer</figcaption></figure><p>I finished editing the video at 5:00 AM, and at 4:30 PM that same Friday, it was played in our company-wide Friday meeting (named WTF, for Welcome to Friday). Everyone on Zoom was encouraged to wear red, and the entire Google Slide deck was given a Lunar New Year refresh. I felt touched that I was able to share a piece of our culture with everyone. The messages and greetings pouring in from the chat, celebrating with us, despite coming from different cultures, genuinely warmed my heart.</p><p>For the first time in a while, although I was miles away from home, I felt like I had, in my own little way, celebrated Lunar New Year.</p><p><em>Gong Hei Fatt Choi, Chúc mừng năm mới !</em></p><h4>Presenting… Seer Lunar New Year’s special</h4><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FO61IhXPNK1I%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DO61IhXPNK1I&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FO61IhXPNK1I%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/a5293db4f54d7b70812ee897f8d1887f/href">https://medium.com/media/a5293db4f54d7b70812ee897f8d1887f/href</a></iframe><p>If you enjoyed this post and want to be part of Seer’s growing culture (and change the world while at it), <a href="https://seermedical.com/careers/">reach out to us. </a>We’re always on the lookout for good humans 🌈</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/281/1*RQZ6DJXC2_KkGWbG5p0wFg.gif" /><figcaption>The author in her taboo cheongsam</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d9c0fae84e4a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seer-medical/a-very-seer-lunar-new-year-d9c0fae84e4a">A very Seer Lunar New Year</a> was originally published in <a href="https://medium.com/seer-medical">Brain Waves</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Adapting computer vision models for time series and automated EEG review]]></title>
            <link>https://medium.com/seer-medical/adapting-computer-vision-models-for-time-series-and-automated-eeg-review-66ff20c1bf9?source=rss----19591ed385e2---4</link>
            <guid isPermaLink="false">https://medium.com/p/66ff20c1bf9</guid>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[eeg]]></category>
            <category><![CDATA[convolutional-network]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Tim Esler]]></dc:creator>
            <pubDate>Tue, 15 Feb 2022 04:42:15 GMT</pubDate>
            <atom:updated>2022-02-15T04:42:15.127Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Tricks for applying state-of-the-art computer vision models to time series and EEG analysis.</em></p><h3>Assisting diagnosis with deep learning</h3><p>For diseases like epilepsy, collection of physiological time series data (such as EEG and ECG) is vital in making diagnoses and directing treatment. However, these data often represent huge amounts of information (a single 7-day EEG study represents over 3 billion data points!) and require expert reviewers to comb through it looking for areas of concern. This can be an incredibly time-consuming process, making diagnosis a long and potentially expensive process.</p><p>To overcome these issues, Seer employs deep learning techniques aimed at highlighting clinically relevant regions in time series signals such as EEGs. Importantly, the clinician remains in the loop and makes all clinical decisions, but by quickly drawing their attention to the interesting parts of a study (some of which are genuinely difficult to find even for trained experts), the process is substantially faster and more effective.</p><blockquote>By highlighting regions of clinical significance, deep learning-assisted EEG analysis can reduce review time by 5–100x.</blockquote><p>In this blog post, I will describe some tricks we use at Seer for building assisted EEG analysis tools using deep learning. Specifically, I will demonstrate methods for translating many state-of-the-art convolutional neural networks (CNNs) from computer vision into the domain of EEG analysis and time series analysis more broadly.</p><h3>Convenient advances in deep learning</h3><p>In recent years, massive amounts of work (not to mention time, money, data, competitions, PhD theses, etc) have been thrown at building amazing deep learning techniques for a range of different applications. While a portion of this has been directed specifically at EEG applications and other closely related time series analysis (e.g., ECG), the vast majority has focused on applications in computer vision, natural language processing, and audio processing — basically, the things that AI giants like Google/Microsoft/Nvidia/OpenAI/etc. have turned their attention to.</p><p>Fortunately, many of the advancements in these seemingly unrelated fields can be translated readily into the field of time series analysis, and specifically automated EEG review, with some minor modifications to their architecture.</p><p>Consider a ResNet model, which is a common CNN used for image classification, and also forms the backbone of many object detection and segmentation models. It operates on image data represented by [width in pixels ×height in pixels ×number of colour channels]. In contrast, EEG data is represented by [length in time points ×number of signal channels]. Hence, EEG channels are composed of 1 dimension (time) whereas images have 2 (width and height), and by mapping 2D operations to 1D operations, it’s possible to apply the same model architecture to both data types.</p><blockquote>Translation of image models into 1D models enables us to take advantage of decades of breakthroughs in computer vision in solving time series problems.</blockquote><h3>Translating vision models to time series</h3><p>There are two main ways we can take advantage of a computer vision model for time series. The first is to modify the deep learning architecture to work with EEG data, and the second is to modify the EEG data to be more like image data. This blog will deal with the former, outlining the approach using the example of a ResNet9 image classification model.</p><p>The majority of deep convolutional models for computer vision consist of only a few different neural network layers. These include 2D convolutions, batch normalisation, max pooling, average pooling, and nonlinearity/activation layers (often ReLU, or rectified linear unit). In modern deep learning frameworks like PyTorch, since each of these layers has a 1D implementation as well, conversion to a time series architecture is usually relatively straightforward.</p><p>Consider the following torch.nn.Conv2d PyTorch layer:</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/d2cfb8000846250b34c34412f6531798/href">https://medium.com/media/d2cfb8000846250b34c34412f6531798/href</a></iframe><p>How do we convert the above 2D convolutional layer to a 1D convolutional layer? Very easily — we can just change one character and use PyTorch’s torch.nn.Conv1d layer instead. There are, of course, more complicated cases than this involving non-square convolutional kernels, striding, and input padding, but these can also be quickly reconciled with an equivalent 1D layer.</p><p>Before jumping into the code and modifying a 2D CNN for time series data, there’s a few things we need to consider. These include (1) the size of our convolutional kernels (and the <em>receptive field</em>), (2) the number of input channels in our data, and (3) the number of output nodes (i.e., classes for a classification model) we need.</p><h4>1. The model’s receptive field</h4><p>Commonly, an important consideration when translating a 2D CNN to a 1D CNN is the size of the convolutional kernels and the model’s <em>receptive field</em> at different parts of its depth. The receptive field corresponds to the region of the input data that contributes to each point in the output, demonstrated for a single layer here:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/272/1*KkGP9oJEUNv9YUecPQ1uKQ.gif" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/351/1*PEDdWuzwB15CPd-cUhTdwQ.gif" /></figure><p>Convolutional layers with larger kernels have larger receptive fields because they draw on larger regions of the inputs in generating outputs. As layers are stacked together, the receptive field of the model grows, as shown below. For applications in audio processing and, to a lesser extent, EEG and ECG analysis, it is typical to use larger kernels than in computer vision, due to high data sampling rates and processing speed considerations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Ah0OVgVf0Rs52x-V" /></figure><p>The chosen kernel size for each of the layers in a model, and the corresponding receptive field, should generally be determined based on the time series problem at hand. Let’s say an image model has been optimised for a receptive field of 21x21 pixels using a series of 10 consecutive convolutional layers with 3x3 kernels. If this model was translated to a 1D CNN for EEG, a larger receptive field is warranted as many EEG morphologies require more than 21 time points to be recognised. If we instead use 10 kernels of size 7, we have a receptive field of 61, allowing the model to recognise patterns in the input that span 61 time points (or 0.24 seconds at 256 Hz). There are a few other tricks to increasing the receptive field, such as using kernel dilation — if you are interested in learning more about receptive field arithmetic, there’s <em>lots</em> to be read about it; <a href="https://theaisummer.com/receptive-field/">here’s one example</a>.</p><h4>2. Input channels</h4><p>A very common requirement when translating deep learning models from one problem to another is modification of the input and output layers to accommodate the shape of the input data and desired output labels. For the vast majority of computer vision models, inputs are fed in with 3 colour channels, usually RGB. This means that an input image with 256x256 pixels is represented by data with a shape of 3x256x256. Time series data typically has a different number of input channels, with audio commonly having 1 (mono) or 2 (stereo) channels. For standard “<a href="https://en.wikipedia.org/wiki/10%E2%80%9320_system_(EEG)">10–20</a>” EEG data, there are 21 channels. So for a 10 second snippet of EEG data sampled at 256 Hz, a single input would be represented by 21x2560 data points. Changing a model to accommodate a different number of input channels is usually an easy task and requires only modification of the first layer in the deep learning network. Note that, for simplicity, each of the figures above shows only a single channel for each convolution. In reality there can be an arbitrary number of channels that differ between inputs and outputs.</p><h4>3. Output classes</h4><p>Similarly, when applying any model to a problem that it wasn’t first trained on, the new problem often requires a different number of outputs from the model. Take image detection for example. If you trained a model on the 600 classes of the <a href="https://storage.googleapis.com/openimages/web/index.html">Open Images detection dataset</a> and wanted to fine-tune the model on the 80-class <a href="https://cocodataset.org/#home">COCO dataset</a>, the number of output nodes on the model’s final layer would have to be modified from 600 to 80. This holds when translating a computer vision model to a time series model. In the case of image classifiers, it’s (very) common for them to have been trained on the classic <a href="https://www.image-net.org/">ImageNet dataset</a> with 1000 classes. Automated EEG analysis, in contrast, may require only a few classes (e.g., interictal, seizure, artefact, etc).</p><p>Due to the frequency with which input channels and output classes are changed, it’s often good to parameterise these so they can be passed to your model by the user, as has been done in the code snippet below.</p><h3>Putting it all together</h3><p>Finally, let’s put all of that into action and convert a 2D image classifier into a 1D time series classifier. Below is a complete example showing how to modify a ResNet9 model, while also increasing the receptive field size by lengthening our kernels.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/0dece7062e7c43e7fe6add758f708cec/href">https://medium.com/media/0dece7062e7c43e7fe6add758f708cec/href</a></iframe><p>To summarise those changes, by changing only 4 lines of code, we have translated a 2D ResNet9 model into a time series classifier and increased the receptive field of its convolutional layers. The beauty of this is that, once you get the hang of it, despite this being a relatively simple example, the same approach can be applied to the vast majority of deep convolutional models, including most of the models from many popular open-source deep computer vision libraries.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=66ff20c1bf9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seer-medical/adapting-computer-vision-models-for-time-series-and-automated-eeg-review-66ff20c1bf9">Adapting computer vision models for time series and automated EEG review</a> was originally published in <a href="https://medium.com/seer-medical">Brain Waves</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building animations in React Native with accessibility in mind]]></title>
            <link>https://medium.com/seer-medical/building-animations-in-react-native-with-accessibility-in-mind-738a6bd86603?source=rss----19591ed385e2---4</link>
            <guid isPermaLink="false">https://medium.com/p/738a6bd86603</guid>
            <category><![CDATA[react]]></category>
            <category><![CDATA[health-technology]]></category>
            <category><![CDATA[animation]]></category>
            <category><![CDATA[react-native]]></category>
            <category><![CDATA[accessibility]]></category>
            <dc:creator><![CDATA[Marton Bodonyi]]></dc:creator>
            <pubDate>Mon, 31 Jan 2022 22:23:42 GMT</pubDate>
            <atom:updated>2022-01-31T22:23:41.948Z</atom:updated>
            <content:encoded><![CDATA[<blockquote><em>The Seer Mobile app helps people with epilepsy manage their seizures, track medications and have better conversations with their doctor.</em></blockquote><p>Something that often gets overlooked when developing animations in apps is accessibility. I get it. You start by going down the rabbit hole of building complex animations (<em>fun!</em>), combining those animations with gestures (<em>less fun</em>), then trying to link that to react-navigation (<em>“when will this end?”</em>) and you’ve got yourself a pretty terrifying looking PR. Don’t get me started on unit tests.</p><figure><img alt="Pull request toolbar showing 22 changed files" src="https://cdn-images-1.medium.com/max/653/1*NLDwQZzNZtYoZP4ykCbahA@2x.png" /><figcaption>What have I done?</figcaption></figure><p>In all that chaos of building complex animations, it’s easy to forget to implement accessibility handlers to ensure people with <em>‘</em><strong><em>Reduce Motion’</em></strong> switched on receive a different experience. <em>Reduce Motion</em> <strong>doesn’t mean cut all animations.</strong> If you look at the way it’s implemented across the iOS and Android system stock apps, subtle animations like fades are all still maintained, but scale and transform animations get cut.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/830/1*GjGYCYp6sM0i7dQsmWb6tQ.gif" /><figcaption>On the left, default motion settings on iOS. On the right, ‘Reduce Motion’ enabled.</figcaption></figure><h3>Why it’s important</h3><p>Let’s get something straight, I love animations. If there was an option in settings called <em>‘Increase Motion’</em> I’d probably tick that but as much as animations bring joy and magic to user interfaces for many users — they can do the exact opposite for some. Around <a href="https://www.epilepsy.com/learn/triggers-seizures/photosensitivity-and-seizures">3% of people with epilepsy can have seizures triggered by flashing lights and visual patterns</a>, but users can opt to <em>Reduce Motion</em> for <a href="https://alistapart.com/article/designing-safer-web-animation-for-motion-sensitivity/">other</a> <a href="https://source.opennews.org/articles/motion-sick/">reasons</a> too.</p><h3>Enabling on device</h3><p>Thankfully, unlike voiceover, you can enable reduce motion from within iOS simulator and Android emulators.</p><h4>On iOS</h4><p>On iOS it’s pretty easy to enable reduce motion. Jump into the accessibility menu, tap ‘Motion’ and check ‘Reduce Motion’.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/390/1*fr3f2V9UD3g0ASdzv1cGwQ.gif" /><figcaption>On iOS: Settings -&gt; Accessibility -&gt; Motion -&gt; Reduce Motion</figcaption></figure><p>On Android, frustratingly the ‘Remove animations’ accessibility option in Settings doesn’t get picked up by React Native. I haven’t been able to find much documentation on how to access that using Android native code but I did find this gem in the React Native documentation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/825/1*YmvLBMgRqO3-3PjKMGDJuQ@2x.png" /></figure><p>So basically, on Android it’s a developer setting. And to access it these are the steps you (or a user) would need to take…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/514/1*39WTFWF19bj2FPAsyF6Wig.gif" /></figure><p>There’s an open issue talking about this over at <a href="https://github.com/facebook/react-native/issues/31221">https://github.com/facebook/react-native/issues/31221</a>, if you know of a way to implement reacting to ‘Remove animations’ in native Java code I’d be happy to help patch React Native to get it in there.</p><h3>Detecting ‘<em>Reduce Motion</em>’ using a hook</h3><p>In React Native, ‘AccessibilityInfo’ has an event listener and async method for querying reduce motion which can be neatly bundled into a hook to use in animations.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/5db371705d22a535cf34287e2544988b/href">https://medium.com/media/5db371705d22a535cf34287e2544988b/href</a></iframe><p>You can also use <a href="https://github.com/infiniteluke/react-reduce-motion">https://github.com/infiniteluke/react-reduce-motion</a> but for simple hooks like this I think it’s better to maintain the code yourself, especially if you’re using TypeScript.</p><h3>Wiring it up to an animation</h3><p>The easiest way to reduce the motion of an animation you’ve implemented is to find which styles have translates in them and swap them out for opacity transitions. It’s neat and can hook in with gestures pretty easily but it comes with the limitation of needing to use a constant animation duration across all your animated styles.</p><p>Here’s some example code that implements the hook we created above.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/bc1dece805f5d4779fa7ba8d23418ba4/href">https://medium.com/media/bc1dece805f5d4779fa7ba8d23418ba4/href</a></iframe><p>Here’s what it looks like side by side in the simulator.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/810/1*3xetTgCVyEak-Xj6urJstA.gif" /><figcaption>On the left, the popup scales using the default motion settings. On the right, the popup fades with Reduce Motion enabled.</figcaption></figure><h3>Support in open source libraries</h3><p><em>Reduce Motion</em> support in popular React Native libraries like <a href="https://github.com/react-navigation/react-navigation/search?q=reduceMotionChanged">react-navigation</a> is spotty at the moment so while you’re implementing it in your own projects, there’s an awesome opportunity to get some high-profile open-source PR’s in and make React Native more accessible in the process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cX-S1V6IlH-XHQSUaOWgTg@2x.png" /></figure><p>You can check out the useReduceMotion hook at <a href="https://gist.github.com/martsie/4f3bf6e3c67cf032ba562745a9c161af">https://gist.github.com/martsie/4f3bf6e3c67cf032ba562745a9c161af</a> and example implementation at <a href="https://gist.github.com/martsie/a09b8192fe9ad40afc1a26ffc1c3c81c">https://gist.github.com/martsie/a09b8192fe9ad40afc1a26ffc1c3c81c</a>.</p><p>If you’re interested in working on delightful, accessible apps built in modern tech that genuinely improves the lives of people with epilepsy, <a href="https://seermedical.com/careers/">come work with us at Seer</a>!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=738a6bd86603" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seer-medical/building-animations-in-react-native-with-accessibility-in-mind-738a6bd86603">Building animations in React Native with accessibility in mind</a> was originally published in <a href="https://medium.com/seer-medical">Brain Waves</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Testing React Context Providers using React Native Testing Library]]></title>
            <link>https://medium.com/seer-medical/testing-react-context-providers-using-react-native-testing-library-7f6823c2d43a?source=rss----19591ed385e2---4</link>
            <guid isPermaLink="false">https://medium.com/p/7f6823c2d43a</guid>
            <category><![CDATA[react-testing-library]]></category>
            <category><![CDATA[jest]]></category>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[react]]></category>
            <category><![CDATA[react-native]]></category>
            <dc:creator><![CDATA[Marton Bodonyi]]></dc:creator>
            <pubDate>Tue, 18 Jan 2022 01:00:27 GMT</pubDate>
            <atom:updated>2022-01-18T01:00:27.500Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>The Seer Mobile app helps people with epilepsy manage their seizures, track medications and have better conversations with their doctor.</blockquote><p>At Seer we use <a href="https://testing-library.com/docs/react-native-testing-library/intro/">React Native Testing Library</a><em> </em>to test the <a href="https://seermedical.com/seer-app/">Seer Mobile App</a> and recently we hit some issues trying to test context provider value assignment in an isolated way. It’s easy to test the consumption of context values in React Components by <a href="https://testing-library.com/docs/example-react-context/">rendering your own provider in a test case</a> but how do you test that a context provider gets assigned the correct values in a component that’s rendering them?</p><h3>Wait, why would I need to test a Context Provider?</h3><p>For the Seer Mobile App, we use contexts to <a href="https://reactnavigation.org/docs/5.x/nesting-navigators#each-screen-in-a-navigator-has-its-own-params">pass params down through nested navigator stacks</a>. When all is said and done, the flow looks a little something like this.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vM44S6rvVA24AEetkbe9mA.jpeg" /></figure><p>Testing the child screens of the navigator is easy, each screen gets its own test that re-implements the context provider with various values. When it comes time to test NestedNavigator though, that’s when things get tricky. How do we know that NestedNavigator is assigning the params correctly to the context provider? One way to to test might be to mock ChildScreenA and override it with some text elements that render out the entire context value — but sometimes our params are pretty complex. The other way might be to write an end-to-end test that runs the entire flow as a user would see it and makes sure the context values are displayed as is. Clean but fragile because if a child screen gets updated then the end-to-end test will fail, even though the only thing we want to test is that context values are available to any child screen that needs them. Fear not, there is a way…</p><h3>Spying on context provider values</h3><p>Unfortunately context providers in React can’t be passed directly through to jest.spyOn but we can use mocks to get around that by re-implementing Context Provider as a function without changing its behaviour.</p><p>In the example below, we’re mocking our custom context to wrap our React-provided context provider in a jest mocked function that can be inspected later by the assertion in our test. The behaviour of the wrapper and the child components remains untouched so that the only thing that’s tested is the Context Provider value injection itself.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/61ac976946feecf73d1a7851248b6413/href">https://medium.com/media/61ac976946feecf73d1a7851248b6413/href</a></iframe><p>Admittedly the purpose of using <em>React Testing Library</em> is to test React components as a user would see them <a href="https://testing-library.com/docs/react-testing-library/intro/#the-problem">as opposed to testing implementation details</a> however in a case like this in which your context provider values can be consumed by many components, in many situations, I think it’s worth bending the rules a little.</p><p>You can check out a working example at <a href="https://github.com/martsie/context-provider-testing">https://github.com/martsie/context-provider-testing</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7f6823c2d43a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seer-medical/testing-react-context-providers-using-react-native-testing-library-7f6823c2d43a">Testing React Context Providers using React Native Testing Library</a> was originally published in <a href="https://medium.com/seer-medical">Brain Waves</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Finding Patterns in Seizure Data]]></title>
            <link>https://medium.com/seer-medical/finding-patterns-in-seizure-data-b1ca8369cbfe?source=rss----19591ed385e2---4</link>
            <guid isPermaLink="false">https://medium.com/p/b1ca8369cbfe</guid>
            <dc:creator><![CDATA[Dominique Eden]]></dc:creator>
            <pubDate>Sun, 28 Nov 2021 21:29:47 GMT</pubDate>
            <atom:updated>2026-06-10T06:53:28.676Z</atom:updated>
            <content:encoded><![CDATA[<p><em>How we use circular statistics to help warn people of a seizure.</em></p><h4><strong>The challenges of uncontrolled seizures</strong></h4><p>Around one third of people with epilepsy are unable to control their seizures with medication. Practically, this means that there are a significant number of people that are living with seemingly random, unprovoked seizures.</p><p>It hardly needs to be explained that living with uncontrolled seizures makes even the most mundane, day-to-day tasks, challenging — pouring hot water, grocery shopping, or swimming are just some examples that’d require careful consideration.</p><p>In 2016, <a href="https://www.epilepsy.com/sites/core/files/atoms/files/community-survey-report-2016%20V2.pdf">the Epilepsy Foundation released a survey</a> asking people about their experience of living with epilepsy. Over 70% said that the <em>unpredictability</em> was the worst part.</p><p>So, if seizures can’t be controlled, can we at least tell when they might happen? In short, yes — we can.</p><p>Providing people with a system that forewarns them of the risk of a seizure would relieve some of the uncertainty associated with epilepsy. Being able to forecast seizures, hours to days in advance, would give someone the opportunity to manage, plan, and respond appropriately. For instance, knowing that one might have a high risk of a seizure in the afternoon could mean a change of plans to the grocery store.</p><blockquote>“Knowing when I might have a seizure would be life-changing. It means I could meaningfully coordinate my day, and my life.” — Epilepsy Patient, 2021</blockquote><h4><strong>But aren’t seizures meant to be random?</strong></h4><p>Seizures are often considered as random. Why? Seizures appear to behave randomly if inspected at a population-level. However, our research as found that seizures can in fact occur at regular intervals when looking at an individual-level, meaning seizures a person-specific. We found that seizures follow rather predictable cycles, recurring at specific times of the day, week, month and/or anything in between.</p><p>The further conclusion to this is that, seizures may not be entirely triggered by external stimuli, but coincide with internal biological mechanisms and rhythms. We’ll save that conversation for another day, however.</p><blockquote>We found that seizures follow rather predictable cycles, recurring at specific times of the day, week, month, and/or anything in between.</blockquote><p>This has been an exciting discovery that has fuelled further research and opened up the possibility to forecasting seizures without specialised equipment or invasive surgery— just a record of times when the seizures occurred.</p><p>Many people with epilepsy already record when their seizures occur, using anything from pen and paper, to calendars, and Excel spreadsheets. In fact, the need to centralise this information was a core motivator for early developments of the Seer app. Since then, we have been able to collect long-term seizure data from an electronic diary, which in turn has allowed us to developing sophisticated methods for identifying event-based seizure cycles.</p><h4><strong>So, how do we forecast seizures?</strong></h4><p><strong>Detecting a user’s seizure cycle or cycles</strong></p><p>The method we currently use to determine persons’ seizure cycle or cycles could be summarised as testing cycles with varying periods, and selecting periods with the strongest — or in other words, statistically significant — alignment to the persons’ recorded seizure times. Keep in mind that most peoples’ seizures are modulated by more than one cycle, so we tend to select several.</p><p>We start by testing an array of cycles against the user’s data, e.g. cycle periods from 6, 12, 18, 24 hours, up to 3 months. To do so, we calculate the phase of each seizure with respect to a given cycle period. The resulting vector length is then calculated for each cycle period. This is called the r-length. The r-length is normalised between 0 and 1, where an r-length of 1 indicates perfect alignment of seizures (i.e. a unimodal distribution), and an r-length of 0 indicates no alignment (i.e. a uniform distribution).</p><p>Let’s use an example using mock seizure times, in place of seizure times which would typically be logged by a user from the Seer app. Here’s a list of seizure times formatted as Unix timestamps (in milliseconds):</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/1ed267abe063c8acdc1a6c868d941e92/href">https://medium.com/media/1ed267abe063c8acdc1a6c868d941e92/href</a></iframe><p>We convert these timestamps to days since epoch as a convenient point of reference. Testing against the following cycle lengths (in days):</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/e0a099867a53d25d445e640b24dc0fdd/href">https://medium.com/media/e0a099867a53d25d445e640b24dc0fdd/href</a></iframe><p>We can calculate r-lengths and phases (angles) using the following function:</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/c899fe9ae946a78e7d81035f6fcae9d1/href">https://medium.com/media/c899fe9ae946a78e7d81035f6fcae9d1/href</a></iframe><p>Using matplotlib to plot the result r-lengths for the tested cycles, we’d see:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_inC8j_fCE5AtpahfxZaIQ.png" /></figure><p>From this we select the strongest cycle periods indicated by cycle periods with peak r-length. In this example, the mock user has a strong cycle at 24 hours, 36 hours, and at 3 days. Typically, a user will have two or three cycles significant enough to model the behaviour of their seizures.</p><p><strong>Determining a user’s probability of a seizure</strong></p><p>Sine waves are constructed using the period from each of the selected cycles. Seizures are then mapped to the phase of each period and a histogram is computed using about 20 bins.</p><p>Seizures are then fit to a von Mises distribution (similar to a Gaussian distribution but using polar coordinates), where the peak of the distribution — or probability — indicates when a seizure is most likely to occur, and the trough indicates when seizure is least likely to occur. Odds probability is then used to combine the von Mises distributions.</p><p>The result? Here’s a plot generated from the mock user’s data, displaying combined likelihood with a 24 hour cycle and 3 day cycle detected, and projected 60-days into the future.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ONLGSt5ZD78WrXDTcF3AEw.png" /><figcaption>The probability of a user experiencing a seizure using mock data, projected 60 days into the future.</figcaption></figure><p>How did we go? Well, if we use the same forecast, project it into the past, and map the mock user’s seizure times, we see good alignment just by visual inspection (see below). Most seizures have occurred at, or very close to, the peak of seizure probability.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iZImRCzQBONG9KQaMmKLoA.png" /><figcaption>The probability of a user experiencing a seizure, projected 60 days into the past. The user’s mock data is indicated by the red triangles, which by visual inspection, show very close alignment to peak probability.</figcaption></figure><p>Despite the fact that this is mock data, <a href="https://onlinelibrary.wiley.com/doi/abs/10.1111/epi.16485">the method has very promising results for real data</a> too.</p><p><strong>Optimising to define a user’s seizure risk</strong></p><p>An optimisation step is applied to provide the user with three levels of risk: high, medium, and low. As a general rule, we opt to have a higher percentage of seizures in high risk, and a higher proportion of time in low risk.</p><h4><strong>Why don’t we use Machine Learning?</strong></h4><p>At some point you may have wondered why we don’t use an out-of-the-box machine learning approach. First and foremost, the statistical method explained above straight up outperforms machine learning models thus far. Another reason, is that machine learning is not suited to just a few data points which is important considering new users and those that may only experience a handful of seizures a year will not have extensive data sets. Finally, being that each model is patient-specific, one model per user and frequent retraining become computationally expensive.</p><p>This isn’t to say however that there is no room for machine learning at all. More recently, we have integrated heart rate from wearable devices to improve forecasting performance, <a href="https://www.frontiersin.org/articles/10.3389/fneur.2021.704060/full">with neural networks showing promise</a>.</p><h4><strong>How this helps people</strong></h4><p>We’re now at a stage where seizures logged by users in the Seer app can be used to detect user-specific seizure cycles and probability, and seizure risk can be determined and displayed back in the app.</p><p>We hope that by alleviating the uncertainty of seizures for those living with epilepsy, we can improve the lives of people with epilepsy.</p><p><em>Want to learn more? Check out our </em><a href="https://seermedical.com/research-and-publications/"><em>research publications page</em></a><em> for some academic deep-dives into these topics.</em></p><p><em>This blog series may contain “work in progress” technological developments and it is meant to be a developer focused forum. As such, posts here are not intended to make any claims relating to clinical safety / efficacy of already cleared medical devices at Seer. The discussion is general in nature and does not include any personally identifiable or health information.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b1ca8369cbfe" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seer-medical/finding-patterns-in-seizure-data-b1ca8369cbfe">Finding Patterns in Seizure Data</a> was originally published in <a href="https://medium.com/seer-medical">Brain Waves</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The challenges & impacts of engineering for epilepsy]]></title>
            <link>https://medium.com/seer-medical/the-challenges-impacts-of-engineering-for-epilepsy-5b44be87516d?source=rss----19591ed385e2---4</link>
            <guid isPermaLink="false">https://medium.com/p/5b44be87516d</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[sofware-development]]></category>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[epilepsy]]></category>
            <dc:creator><![CDATA[Josh Reich]]></dc:creator>
            <pubDate>Fri, 22 Oct 2021 03:37:17 GMT</pubDate>
            <atom:updated>2021-10-22T03:37:17.832Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>Hi, I’m Josh Reich, the CTO at Seer Medical, an Australian medical technology company specialising in the diagnosis and management of neurological disorders. I lead the teams building our software. In launching this blog, we’re hoping to highlight the work that our team is doing to use advanced technologies, from material science and electronics to software and interaction design to have a deeply meaningful positive impact on the lives of people with epilepsy. We have plans to dive into a range of topics and share our work with the broader technology community.</blockquote><h3><strong>A very short introduction to the electrophysiology of epileptic seizures.</strong></h3><p>Epilepsy is a disorder in the brain that impacts about 1% of the world’s population, with about 1 in 26 people experiencing a seizure at some point in their lives, and about 50 million people currently diagnosed. For people who suffer epileptic seizures, their lives are punctuated by seemingly random events that can range from momentary loss of awareness, localised motor or speech disorders, through to the stereotyped tonic-clonic seizures, with loss of consciousness, muscle stiffness and violent shaking.</p><p>While there are a range of underlying causes, with roughly one third of folks having their first seizure in childhood, and others developing it later in life, epilepsy is characterised by spreading waves of electrical depolarisation in neurons. As different regions of the brain are impacted, their function is disrupted. This makes an electroencephalograph (EEG) a vital tool in diagnosing and localising the disease.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GUtc6wnZFzhk0Lgm" /><figcaption>Stylised neurons connecting in the brain</figcaption></figure><p>The function of neurons is complex. At the large cell body, dendrites reach out to nearby neurons where they meet at a synapse. Neurotransmitters, released into the synapse, bind with receptors on the dendrites and when the right combination and volume of binding occurs, it triggers an action potential down the long axon of the neuron. This action potential is a voltage that develops across the cell membrane, induced by the activity of ion channels that allow charged ions to cross the membrane. This voltage wave travels down the axon and then triggers the release of neurotransmitters at the synaptic terminals where further neurons are impacted.</p><p>In normal brain function, billions of neurons are processing information in a relatively uncoordinated fashion. This results in a near net-zero charge measurable on the scalp. During a seizure synchronisation between neurons occurs, and voltages in the order of 10’s of microvolts can be measured by an EEG.</p><p>To diagnose epilepsy, people with seizures are typically referred to a hospital setting where they are connected to an EEG and monitored by clinical staff around the clock for seven days. In Australia, patients can wait anywhere from six to 18 months for a diagnostic session. Waiting upwards of a year - while having seizures, while losing consciousness, while living with a suspected disease affecting their brain. Unnecessary, anxious waiting, just to find out what’s going on.</p><p>In 2017 our founders believed that there was a better way. A small group of neurologists, computational neuroscientists and electrical engineers got together to build Seer Medical with the aim of giving patients faster access to accurate diagnosis - from the comfort of their own homes.</p><h3><strong>Introducing Seer Medical</strong></h3><p>Seer has tackled the issues of accessibility and long wait times for hospital monitoring through two critical innovations. We’ve taken equipment that historically needed to be wheeled around hospitals on a cart and always plugged into the wall outlet and miniaturised it to a small wearable device. With low power analog to digital converters (ADCs), microcontrollers and radios, a patient can be fitted with electrodes in one of our offices, and can wear the device for 10 days without a recharge.</p><p>Redesigning the hardware from the ground up allowed us to take a patient-centric approach, so we’ve built our own magnetic EEG connector, that places the ADCs directly on the head, lowering noise, and allowing the patient to quickly remove it to shower — something they can’t do at all easily in hospital. We&#39;ve also invented a new water-based electrode adhesive (WaterTabs®) to replace collodion, an inflammable noxious solution that is one of the leading causes of complaints with traditional EEGs. WaterTabs are innocuous and non-irritant, and can be simply washed off at the end of a study.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*wfanEwdQ5CqpwIDa" /><figcaption><em>Screenshot of the Seer Cloud application. Patient data provided with consent.</em></figcaption></figure><p>The other main innovation is our use of machine learning to interpret EEG signals. To the untrained eye (myself included), EEG traces look like squiggly lines. I’ve sat in frequently with neurologists as they review patient data. They have an uncanny ability to see through artefacts from patients chewing, head movement, and other sources of noise to find clear signs of the synchronised depolarisation that characterises epilepsy. For large seizures affecting the entire brain, these can be easy to identify. For smaller localised epileptiform events, the signs on the EEG are more subtle. Over the course of their training and careers, they can read the charts as easily as the rest of us can read writing, or musicians can read a score.</p><p>We’ve found that deep learning techniques from computer vision and audio processing — namely image segmentation, object detection, and audio event detection — can be readily translated into EEG applications. In recent years, these fields have experienced an explosion in cutting-edge research, yielding rapid improvement in automatic techniques for identifying complex patterns in data. In turn, this has yielded tremendous benefit for translatable applications, such as EEG analysis. When combined with large, well-curated, expert-verified EEG datasets, these techniques are able to identify regions of interest as well as a trained neurologist. At Seer, we’ve been successful in combining the best bits from these different fields to build state-of-the-art analysis techniques. And unlike many tech companies touting ML, we’ve demonstrated that our systems are clinically effective and are approved for use in many countries around the world.</p><p>This is a critical enabler for at-home epilepsy diagnosis. In a hospital setting, a clinician is constantly monitoring the patient. They’re watching the patient’s behaviour and observing the EEG traces. In addition to the cost of the bed, the 24/7 clinical presence is a significant driver of the cost of diagnosing the disease. Hospitals have more profitable uses of their facilities, so there is a limit to the number of patients who can be diagnosed. And with COVID-19, more patients are hesitant to spend extended periods in hospital settings. Our ML system means that instead of having to spend days watching a replay of the EEG, we can quickly direct our clinical staff to only review the relevant time windows, reducing the cost and time to diagnosis by at least one order of magnitude.</p><p>Our ML models are also starting to change the way that we can present EEG data. While neurologists who specialise in EEG reading are fluent with current visualisations, we believe that we can help them do their jobs faster and gain a deeper understanding of each patient with unique visualisations of the data that we surface. This is an interesting design challenge, and if we do it well, we’ll also be able to help patients and their primary carers get better insight into the nature of their disease.</p><p>Behind the scenes, we’ve deployed all of this in a compliant fashion on the AWS cloud infrastructure. Each patient study produces gigabytes of time series electrode data that is synchronised with many more gigabytes of high resolution video. The video allows clinicians to correlate what they’re seeing on the traces with patient behaviour. Given that we do over twice as many EEG studies as all hospitals in Australia, combined, our data set is now over 200TB. Our devops approach enables us to work quickly as we enhance our systems, unlocking the unique approach we take to analysing EEG data to enable more accurate diagnosis, faster, and cheaper.</p><h3>Seizure prediction and challenging commonly accepted knowledge</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J-sNE2fiom_a2N-IgA5FZQ.png" /><figcaption>Figure from 2016 Epilepsy Foundation Community Survey</figcaption></figure><p>If you look to the world&#39;s most commonly consulted source of information - <a href="https://en.wikipedia.org/wiki/Epilepsy#Seizure_prediction_and_modeling">Wikipedia</a> - it will tell you that there are no effective ways of predicting seizures. The <a href="https://www.epilepsy.com/sites/core/files/atoms/files/community-survey-report-2016%20V2.pdf">2016 Epilepsy Foundation Community Survey Report</a> reveals that the leading impact of the disease on patient’s lives is its unpredictability. And yet we have found that for individual patients the <a href="https://www.seermedical.com/research-and-publications/">risk of seizures are predictable</a>. We&#39;re proving that the common understanding of the disease is wrong. Not only does this allow us better schedule EEGs to capture seizures, but the positive impact on patients’ lives is difficult to overstate.</p><p>There’s a common misconception that for most patients seizures are triggered by well observed events (e.g., flashing lights). While this is true for a minority of folks, many live each day not knowing what the day brings. This can be an extreme burden. We’re currently trialing these forecasts via our mobile app with a range of patients, and the results are giving us hope that we can help people with uncontrolled seizures, and empower them to live their lives more freely.</p><p>—</p><p>In my prior career, I was the CEO of an online banking company that was designed to help people live their financial lives with more confidence. We put our customers at the centre of our work and used design and technology to build a system for how people think rather than how banks work. Our work was rewarding and we built a community of fans who were able to develop a more healthy relationship with money to better achieve their financial goals.</p><p>As worthwhile as that work was, it pales in comparison to the impact we’re having here at Seer. Each day I’m humbled by the brilliance of the people across our teams and the stories we hear from our patients and those in the broader epilepsy community. I’m looking forward to sharing more of our stories and building upon our multidisciplinary approach by publishing our technical findings, challenges, and their connection to the deeply meaningful impact on patients’ lives. My hope is that by being more public with our work, we can also learn from you! If there are specific topics that you’d like us to dive into, or you just want to chat, please feel free to <a href="/service/mailto:josh.reich@seermedical.com">reach out to me</a>.</p><p><em>This blog series may contain &quot;work in progress&quot; technological developments and it is meant to be a developer focused forum. As such, posts here are not intended to make any claims relating to clinical safety / efficacy of already cleared medical devices at Seer Medical. The discussion is general in nature and does not include any personally identifiable or health information.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5b44be87516d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/seer-medical/the-challenges-impacts-of-engineering-for-epilepsy-5b44be87516d">The challenges &amp; impacts of engineering for epilepsy</a> was originally published in <a href="https://medium.com/seer-medical">Brain Waves</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>