{"681793":{"#nid":"681793","#data":{"type":"news","title":"Fill-in-the-Blank Training Primes AI to Interpret Health Data From Smartwatches, Fitness\u00a0Trackers","body":[{"value":"\u003Cdiv class=\u0022theconversation-article-body\u0022\u003E\u003Cp\u003EThe human body constantly generates a variety of signals that can be measured from outside the body with \u003Ca href=\u0022https:\/\/doi.org\/10.2196\/35684\u0022\u003Ewearable devices\u003C\/a\u003E. These bio-signals \u2013 ranging from heart rate to sleep state and blood oxygen levels \u2013 can indicate whether someone is \u003Ca href=\u0022https:\/\/doi.org\/10.1038\/s41746-024-01333-z\u0022\u003Ehaving mood swings\u003C\/a\u003E or can be used to diagnose a variety of \u003Ca href=\u0022https:\/\/www.brighamandwomens.org\/medical-resources\/emg-test#:%7E:text=An%20EMG%20test%20may%20be,by%20pain%20or%20psychological%20reasons.\u0022\u003Ebody\u003C\/a\u003E or \u003Ca href=\u0022https:\/\/www.mayoclinic.org\/tests-procedures\/eeg\/about\/pac-20393875#:%7E:text=An%20EEG%20records%20the%20electrical,electrical%20activity%20in%20the%20brain.\u0022\u003Ebrain disorders\u003C\/a\u003E.\u003C\/p\u003E\u003Cp\u003EIt can be relatively cheap to gather a lot of bio-signal data. Researchers can organize a study and ask participants to use a wearable device akin to a smartwatch for a few days. However, to teach a machine learning algorithm to find a relationship between a specific bio-signal and a health disorder, you first need to teach the algorithm to recognize that disorder. That\u2019s where computer engineers like myself come in.\u003C\/p\u003E\u003Cp\u003EMany commercial smartwatches, such as \u003Ca href=\u0022https:\/\/afibinstitute.com.au\/atrial-fibrillation-a-guide-to-wearable-ecg-smart-watches\/#elementor-toc__heading-anchor-3\u0022\u003Eones by Apple, AliveCor, Google and Samsung\u003C\/a\u003E, currently support atrial fibrillation detection. Atrial fibrillation is a common type of irregular heart rhythm, and leaving it untreated can lead to a stroke. One way to automatically detect \u003Ca href=\u0022https:\/\/my.clevelandclinic.org\/health\/diseases\/16765-atrial-fibrillation-afib\u0022\u003Eatrial fibrillation\u003C\/a\u003E is to train a machine learning algorithm to recognize what atrial fibrillation looks like in the data.\u003C\/p\u003E\u003Cp\u003EThis machine learning approach requires large bio-signal datasets in which instances of atrial fibrillation are labeled. The algorithm can use the labeled instances to learn to recognize a relationship between the bio-signal and atrial fibrillation.\u003C\/p\u003E\u003Cp\u003EThe labeling process can be quite expensive because it requires experts, such as cardiologists, to go through millions of data points and label each instance of atrial fibrillation. The same problem extends to many other bio-signals and disorders.\u003C\/p\u003E\u003Cp\u003ETo resolve this issue, researchers have been developing new ways to train machine learning algorithms with fewer labels. By first training a machine learning model to fill in the blanks of large-scale unlabeled bio-signal data, the machine learning model is primed to learn the relationship between a bio-signal and a disorder with fewer labels. This is called pretraining. Pretraining even helps a machine learning model learn a relationship between a bio-signal and a disorder when it is pretrained on a completely unrelated bio-signal.\u003C\/p\u003E\u003Cfigure class=\u0022align-center zoomable\u0022\u003E\u003Cp\u003E\u003Ca href=\u0022https:\/\/images.theconversation.com\/files\/657861\/original\/file-20250326-57-i0xtcq.png?ixlib=rb-4.1.0\u0026amp;q=45\u0026amp;auto=format\u0026amp;w=1000\u0026amp;fit=clip\u0022\u003E\u003Cimg alt=\u0022A silhouette of a person overlaid with text.\u0022 src=\u0022https:\/\/images.theconversation.com\/files\/657861\/original\/file-20250326-57-i0xtcq.png?ixlib=rb-4.1.0\u0026amp;q=45\u0026amp;auto=format\u0026amp;w=754\u0026amp;fit=clip\u0022 srcset=\u0022https:\/\/images.theconversation.com\/files\/657861\/original\/file-20250326-57-i0xtcq.png?ixlib=rb-4.1.0\u0026amp;q=45\u0026amp;auto=format\u0026amp;w=600\u0026amp;h=453\u0026amp;fit=crop\u0026amp;dpr=1 600w, https:\/\/images.theconversation.com\/files\/657861\/original\/file-20250326-57-i0xtcq.png?ixlib=rb-4.1.0\u0026amp;q=30\u0026amp;auto=format\u0026amp;w=600\u0026amp;h=453\u0026amp;fit=crop\u0026amp;dpr=2 1200w, https:\/\/images.theconversation.com\/files\/657861\/original\/file-20250326-57-i0xtcq.png?ixlib=rb-4.1.0\u0026amp;q=15\u0026amp;auto=format\u0026amp;w=600\u0026amp;h=453\u0026amp;fit=crop\u0026amp;dpr=3 1800w, https:\/\/images.theconversation.com\/files\/657861\/original\/file-20250326-57-i0xtcq.png?ixlib=rb-4.1.0\u0026amp;q=45\u0026amp;auto=format\u0026amp;w=754\u0026amp;h=569\u0026amp;fit=crop\u0026amp;dpr=1 754w, https:\/\/images.theconversation.com\/files\/657861\/original\/file-20250326-57-i0xtcq.png?ixlib=rb-4.1.0\u0026amp;q=30\u0026amp;auto=format\u0026amp;w=754\u0026amp;h=569\u0026amp;fit=crop\u0026amp;dpr=2 1508w, https:\/\/images.theconversation.com\/files\/657861\/original\/file-20250326-57-i0xtcq.png?ixlib=rb-4.1.0\u0026amp;q=15\u0026amp;auto=format\u0026amp;w=754\u0026amp;h=569\u0026amp;fit=crop\u0026amp;dpr=3 2262w\u0022 sizes=\u0022(min-width: 1466px) 754px, (max-width: 599px) 100vw, (min-width: 600px) 600px, 237px\u0022\u003E\u003C\/a\u003E\u003C\/p\u003E\u003Cfigcaption\u003E\u003Cspan class=\u0022caption\u0022\u003EBio-signals are found all over the body and provide information about different bodily functions. Each of these is a bio-signal that measures a specific physiological signal in a noninvasive way.\u003C\/span\u003E \u003Cspan class=\u0022attribution source\u0022\u003EEloy Geenjaar\u003C\/span\u003E\u003C\/figcaption\u003E\u003Cfigcaption\u003E\u0026nbsp;\u003C\/figcaption\u003E\u003C\/figure\u003E\u003Ch2\u003EChallenges of Working With Bio-Signals\u003C\/h2\u003E\u003Cp\u003EFinding relationships between bio-signals and disorders can be difficult because of noise, or irrelevant data, differences between people\u2019s bio-signals, and because the relationship between a bio-signal and disorder may not be clear.\u003C\/p\u003E\u003Cp\u003EFirst, bio-signals contain a lot of noise. For example, when you\u2019re wearing a smartwatch while running, the watch will move around. This causes the sensor for the bio-signal to record at different locations during the run. Since the locations vary across the run, swings in the bio-signal value may now be due to variations in the recording location instead of due to physiological processes.\u003C\/p\u003E\u003Cp\u003ESecond, everyone\u2019s bio-signals are unique. The location of veins, for example, often differ between people. This means that even if smartwatches are worn at exactly the same place on everyone\u2019s wrists, the bio-signal related to those veins is recorded differently from one person to the next. The same underlying signal, such as someone\u2019s heart rate, will lead to different bio-signal values.\u003C\/p\u003E\u003Cp\u003EThe underlying signal itself can also be unique for people or groups of people. The resting heart rate of an average person is around 60-80 beats per minute, but athletes can have resting heart rates \u003Ca href=\u0022https:\/\/www.healthline.com\/health\/athlete-heart-rate\u0022\u003Eas low as 30-40 beats per minute\u003C\/a\u003E.\u003C\/p\u003E\u003Cp\u003ELastly, the relationship between a bio-signal and a disorder is often complex. This means that the disorder is not immediately obvious from looking at the bio-signal.\u003C\/p\u003E\u003Cp\u003EMachine learning algorithms allow researchers to learn from data and account for the complexity, noise and variability of people. By using large bio-signal datasets, machine learning algorithms are able to find clear relationships that apply to everyone.\u003C\/p\u003E\u003Ch2\u003ELearning to Fill in the Blanks\u003C\/h2\u003E\u003Cp\u003EResearchers can use unlabeled bio-signal data as a warmup for the machine learning algorithm. This warmup, or \u003Ca href=\u0022https:\/\/dl.acm.org\/doi\/10.5555\/1756006.1756025\u0022\u003Epre-training\u003C\/a\u003E, primes the machine learning algorithm to find a relationship between the bio-signal and a disorder. This is a bit like walking around a park to get the lay of the land before working out a route to go running.\u003C\/p\u003E\u003Cp\u003EThere are many ways to pretrain a machine learning algorithm. In \u003Ca href=\u0022https:\/\/doi.org\/10.48550\/arXiv.2412.11695\u0022\u003Emy research\u003C\/a\u003E with Dolby Laboratories researcher \u003Ca href=\u0022https:\/\/scholar.google.com\/citations?hl=en\u0026amp;user=EEds7hMAAAAJ\u0026amp;view_op=list_works\u0026amp;sortby=pubdate\u0022\u003ELie Lu\u003C\/a\u003E and \u003Ca href=\u0022https:\/\/doi.org\/10.48550\/arXiv.2309.05927\u0022\u003Eprevious research\u003C\/a\u003E, the machine learning algorithm is taught \u003Ca href=\u0022https:\/\/doi.org\/10.1109\/CVPR52688.2022.01553\u0022\u003Eto fill in the blanks\u003C\/a\u003E.\u003C\/p\u003E\u003Cp\u003ETo do this, we take a bio-signal and artificially create gaps of a certain length \u2013 for example, one second. We then teach the machine learning algorithm to fill in the missing piece of bio-signal. This is possible because the machine learning algorithm sees what the bio-signal looks like before and after the gap.\u003C\/p\u003E\u003Cp\u003EIf the heart rate of a person is around 60 beats per minute before the gap, there will likely be a heartbeat in the one-second gap. In this case, we\u2019re training the machine learning algorithm to predict when that heartbeat will occur.\u003C\/p\u003E\u003Cp\u003EOnce we have trained the machine learning algorithm to do this, it will have found a relationship between someone\u2019s heart rate and when the next beat should occur. We can now train the machine learning algorithm with this relationship between a normal heart rate and bio-signal already learned. This makes it easier for the algorithm to learn the relationship between heart rate and atrial fibrillation. Since atrial fibrillation is characterized by fast and irregular heartbeats, and the algorithm is now good at predicting when a heartbeat will happen, it can quickly learn to detect these irregularities.\u003C\/p\u003E\u003Cfigure class=\u0022align-center zoomable\u0022\u003E\u003Cp\u003E\u003Ca href=\u0022https:\/\/images.theconversation.com\/files\/655466\/original\/file-20250315-56-nfmqu9.png?ixlib=rb-4.1.0\u0026amp;q=45\u0026amp;auto=format\u0026amp;w=1000\u0026amp;fit=clip\u0022\u003E\u003Cimg alt=\u0022three rows of horizontal lines with regularly spaced vertical spikes\u0022 src=\u0022https:\/\/images.theconversation.com\/files\/655466\/original\/file-20250315-56-nfmqu9.png?ixlib=rb-4.1.0\u0026amp;q=45\u0026amp;auto=format\u0026amp;w=754\u0026amp;fit=clip\u0022 srcset=\u0022https:\/\/images.theconversation.com\/files\/655466\/original\/file-20250315-56-nfmqu9.png?ixlib=rb-4.1.0\u0026amp;q=45\u0026amp;auto=format\u0026amp;w=600\u0026amp;h=183\u0026amp;fit=crop\u0026amp;dpr=1 600w, https:\/\/images.theconversation.com\/files\/655466\/original\/file-20250315-56-nfmqu9.png?ixlib=rb-4.1.0\u0026amp;q=30\u0026amp;auto=format\u0026amp;w=600\u0026amp;h=183\u0026amp;fit=crop\u0026amp;dpr=2 1200w, https:\/\/images.theconversation.com\/files\/655466\/original\/file-20250315-56-nfmqu9.png?ixlib=rb-4.1.0\u0026amp;q=15\u0026amp;auto=format\u0026amp;w=600\u0026amp;h=183\u0026amp;fit=crop\u0026amp;dpr=3 1800w, https:\/\/images.theconversation.com\/files\/655466\/original\/file-20250315-56-nfmqu9.png?ixlib=rb-4.1.0\u0026amp;q=45\u0026amp;auto=format\u0026amp;w=754\u0026amp;h=230\u0026amp;fit=crop\u0026amp;dpr=1 754w, https:\/\/images.theconversation.com\/files\/655466\/original\/file-20250315-56-nfmqu9.png?ixlib=rb-4.1.0\u0026amp;q=30\u0026amp;auto=format\u0026amp;w=754\u0026amp;h=230\u0026amp;fit=crop\u0026amp;dpr=2 1508w, https:\/\/images.theconversation.com\/files\/655466\/original\/file-20250315-56-nfmqu9.png?ixlib=rb-4.1.0\u0026amp;q=15\u0026amp;auto=format\u0026amp;w=754\u0026amp;h=230\u0026amp;fit=crop\u0026amp;dpr=3 2262w\u0022 sizes=\u0022(min-width: 1466px) 754px, (max-width: 599px) 100vw, (min-width: 600px) 600px, 237px\u0022\u003E\u003C\/a\u003E\u003C\/p\u003E\u003Cfigcaption\u003E\u003Cspan class=\u0022caption\u0022\u003EMachine learning pre-training on filling in the blanks of a heart bio-signal.\u003C\/span\u003E \u003Cspan class=\u0022attribution source\u0022\u003EEloy Geenjaar\u003C\/span\u003E\u003C\/figcaption\u003E\u003C\/figure\u003E\u003Cp\u003EThe idea of filling in the blanks can be generalized to other bio-signals as well. \u003Ca href=\u0022https:\/\/papers.nips.cc\/paper_files\/paper\/2022\/hash\/194b8dac525581c346e30a2cebe9a369-Abstract-Conference.html\u0022\u003EPrevious research\u003C\/a\u003E \u003Ca href=\u0022https:\/\/iclr.cc\/virtual\/2024\/23539\u0022\u003Ehas shown\u003C\/a\u003E, and \u003Ca href=\u0022https:\/\/doi.org\/10.48550\/arXiv.2412.11695\u0022\u003Eour work\u003C\/a\u003E reconfirmed, that pretraining a model on one bio-signal without any labels allows it to learn clinically useful relationships from other bio-signals with few labels. This shortcut means that researchers can pretrain on bio-signals that are easy to gather and use the machine learning model on ones that are hard to gather and label.\u003C\/p\u003E\u003Ch2\u003EFaster Disorder Detection Development\u003C\/h2\u003E\u003Cp\u003EBy improving pretraining, researchers can make machine learning algorithms better and more efficient at detecting diseases and disorders. Pretraining improvements reduce cost and time spent by experts labeling.\u003C\/p\u003E\u003Cp\u003EA recent example of machine learning algorithms used for early detection is Google\u2019s \u003Ca href=\u0022https:\/\/blog.google\/products\/pixel\/pixel-watch-3-loss-of-pulse-detection\/\u0022\u003ELoss of Pulse\u003C\/a\u003E smartwatch feature. The emerging field of bio-signal pretraining can help enable faster development of similar features using a wider range of bio-signals and for a wider range of disorders.\u003C\/p\u003E\u003Cp\u003EWith increasing types of bio-signals and more data, researchers may be able to discover relationships that dramatically improve early detection of disease and disorders. The earlier many diseases and disorders are found, the better a treatment plan works for patients.\u003C!-- Below is The Conversation\u0027s page counter tag. Please DO NOT REMOVE. --\u003E\u003Cimg style=\u0022border-color:!important;border-style:none;box-shadow:none !important;margin:0 !important;max-height:1px !important;max-width:1px !important;min-height:1px !important;min-width:1px !important;opacity:0 !important;outline:none !important;padding:0 !important;\u0022 src=\u0022https:\/\/counter.theconversation.com\/content\/251890\/count.gif?distributor=republish-lightbox-basic\u0022 alt=\u0022The Conversation\u0022 width=\u00221\u0022 height=\u00221\u0022 referrerpolicy=\u0022no-referrer-when-downgrade\u0022\u003E\u003C!-- End of code. If you don\u0027t see any code above, please get new code from the Advanced tab after you click the republish button. The page counter does not collect any personal data. More info: https:\/\/theconversation.com\/republishing-guidelines --\u003E\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u003Cem\u003EThis article is republished from \u003C\/em\u003E\u003Ca href=\u0022https:\/\/theconversation.com\u0022\u003E\u003Cem\u003EThe Conversation\u003C\/em\u003E\u003C\/a\u003E\u003Cem\u003E under a Creative Commons license. Read the \u003C\/em\u003E\u003Ca href=\u0022https:\/\/theconversation.com\/fill-in-the-blank-training-primes-ai-to-interpret-health-data-from-smartwatches-and-fitness-trackers-251890\u0022\u003E\u003Cem\u003Eoriginal article\u003C\/em\u003E\u003C\/a\u003E\u003Cem\u003E.\u003C\/em\u003E\u003C\/p\u003E\u003C\/div\u003E","summary":"","format":"full_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003EThe human body constantly generates a variety of signals that can be measured from outside the body with wearable devices.\u0026nbsp;\u003C\/p\u003E","format":"limited_html"}],"field_summary_sentence":[{"value":"The human body constantly generates a variety of signals that can be measured from outside the body with wearable devices. "}],"uid":"27469","created_gmt":"2025-04-15 14:06:26","changed_gmt":"2026-03-19 13:16:24","author":"Kristen Bailey","boilerplate_text":"","field_publication":"","field_article_url":"","location":"Atlanta, GA","dateline":{"date":"2025-04-10T00:00:00-04:00","iso_date":"2025-04-10T00:00:00-04:00","tz":"America\/New_York"},"extras":[],"hg_media":{"676841":{"id":"676841","type":"image","title":"AI promises to help wearable devices like smart watches better monitor your health. adamkaz\/E+ via Getty Images","body":"\u003Cp\u003EAI promises to help wearable devices like smart watches better monitor your health. \u003Ca href=\u0022https:\/\/www.gettyimages.com\/detail\/photo\/senior-black-woman-running-with-a-fitness-tracker-royalty-free-image\/1299849508?phrase=smart+watch\u0022\u003Eadamkaz\/E+ via Getty Images\u003C\/a\u003E\u003C\/p\u003E","created":"1744726069","gmt_created":"2025-04-15 14:07:49","changed":"1744726069","gmt_changed":"2025-04-15 14:07:49","alt":"AI promises to help wearable devices like smart watches better monitor your health. adamkaz\/E+ via Getty Images","file":{"fid":"260686","name":"file-20250321-56-l266vi.jpg","image_path":"\/sites\/default\/files\/2025\/04\/15\/file-20250321-56-l266vi.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2025\/04\/15\/file-20250321-56-l266vi.jpg","mime":"image\/jpeg","size":302589,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2025\/04\/15\/file-20250321-56-l266vi.jpg?itok=4GAqIadE"}}},"media_ids":["676841"],"related_links":[{"url":"https:\/\/theconversation.com\/fill-in-the-blank-training-primes-ai-to-interpret-health-data-from-smartwatches-and-fitness-trackers-251890","title":"Read This Article on The Conversation"}],"groups":[{"id":"1237","name":"College of Engineering"},{"id":"658168","name":"Experts"},{"id":"1214","name":"News Room"},{"id":"1188","name":"Research Horizons"},{"id":"1255","name":"School of Electrical and Computer Engineering"}],"categories":[],"keywords":[{"id":"187915","name":"go-researchnews"}],"core_research_areas":[],"news_room_topics":[{"id":"71881","name":"Science and Technology"}],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Ch5\u003EAuthor:\u003C\/h5\u003E\u003Cp\u003E\u003Ca href=\u0022https:\/\/theconversation.com\/profiles\/eloy-geenjaar-2343252\u0022\u003EEloy Geenjaar\u003C\/a\u003E, Ph.D. Student in Electrical Engineering and Computer Engineering, \u003Ca href=\u0022https:\/\/theconversation.com\/institutions\/georgia-institute-of-technology-1310\u0022\u003E\u003Cem\u003EGeorgia Institute of Technology\u003C\/em\u003E\u003C\/a\u003E\u003C\/p\u003E\u003Ch5\u003EMedia Contact:\u003C\/h5\u003E\u003Cp\u003EShelley Wunder-Smith\u003Cbr\u003E\u003Ca href=\u0022mailto:shelley.wunder-smith@research.gatech.edu\u0022\u003Eshelley.wunder-smith@research.gatech.edu\u003C\/a\u003E\u003C\/p\u003E","format":"limited_html"}],"email":[],"slides":[],"orientation":[],"userdata":""}}}