- This compute patch performs instant, on-skin AI analysis for health data
- Minimal latency is critical for conditions like ventricular fibrillation
- It could also overhaul robotics and edge AI for disaster relief
Researchers at the University of Chicago Pritzker School of Molecular Engineering have developed a compute patch that can run AI models directly on the body, rather than sending data to a connected smartphone, cloud server or other external processor.
Published in the Nature Electronics journal, researcher Sihong Wang likened the development to having a “personal, instantaneous doctor integrated into [users’] devices.”
Though it’s far from being commercially available, the tech tackles the fact that most wearables today essentially serve as data collectors only. While smartwatches have long been measuring heart rate, movement, oxygen levels, ECG signals and more, that data typically gets transferred to a smartphone for analysis, or even cloud servers in the case of the newly launched Google Health with AI capabilities.
Skin-based AI inference could revolutionize healthcare
The new patch that’s been developed by researchers performs both the sensing and AI inference directly on the skin, with analysis occurring in milliseconds without relying on wireless communication, cloud computing or other external factors.
Ultra-low latency is especially important for some medical conditions like ventricular fibrillation, where even a few seconds of latency could make a difference.
However, there are other benefits to this technology too, with the paper highlighting a reduction in power consumption and privacy risks thanks to the on-device processing.
Stretchable transistors that bend and conform to the skin are credited with making the patch possible, while conventional chips and rigid silicon in legacy hardware would previously have made this impossible. However, a gel electrolyte layer presented its own challenges, threatening to move like a liquid and short circuit electrical components.
“What we had to ask was whether we could use or change the properties of these polymers to make them compatible with photolithography—the main patterning method used in the microelectronics industry,” Wang added.”
Ventricular fibrillation was clearly a major focus for the study, and thanks to a donated human heart, the team was able to confirm the patch’s ability to locate wavefront positions with 99.6% accuracy.
The research claims to enable the “diverse edge processing functions applicable to various types of health data, including multilayer perceptron (MLP) for heart attack prediction and convolution operations for precise tracking of arrhythmia fibrillation wavefronts on the heart surface.”
While this specific study relates to on-skin patches, it also alludes to a future of “implantable on-body edge” compute for true smart health. “high-resolution signal measurements” are being targeted for next-generation implantables for highly accurate, real-time data from actual living organs.
This technology’s future extends far beyond human healthcare
Besides human data, patches and implantables with integrated compute could also overhaul robotics, giving humanoids human-like senses with real-time accuracy, making them perfect for disaster recovery efforts where wireless communications may be unreliable.
This type of reinforcement learning was put to the test in an ant-like robot study, where the miniature robot was able to navigate environments with comparable success to conventional computer simulations.
Looking ahead, commercial versions of this technology would mark a major shift in artificial intelligence, where the technology gets deployed on the edge instead of in latency-prone data centers.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
https://cdn.mos.cms.futurecdn.net/5W4VKeuChNGM7iQx8ky8JA-1920-80.jpg
Source link
