DiabDeep: Pervasive Diabetes Diagnosis based on Wearable Medical Sensors and Efficient Neural Networks
Diabetes impacts the living quality of millions of people around the globe. Daily diabetes diagnosis and monitoring remain arduous but urgent given that the disease develops and gets treated outside of clinics. The emergence of Wearable Medical Sensors (WMSs) and machine learning points towards a potential way to address this challenge.
Researchers at Princeton University propose a novel framework called DiabDeep that combines efficient neural networks (i.e., DiabNNs) and off-the-shelf WMSs for pervasive diabetes diagnosis. DiabDeep enables both an (i) accurate inference on the server, e.g., a desktop, and (ii) efficient inference on the edge, e.g., a smart phone, to balance accuracy and efficiency based on varying resource budgets and design goals. The inventors have demonstrated the effectiveness of DiabDeep through a detailed analysis of data collected from 52 participants. For server (edge) side inference, they achieve a 96.3% (95.3%) accuracy in classifying diabetics against healthy individuals, and a 95.7% (94.6%) accuracy in distinguishing among type-1 diabetic, type-2 diabetic, and healthy individuals.
When compared to conventional baselines, such as support vector machines with linear and RBF kernels, k-NN, random forest, and linear ridge classifiers, DiabNNs achieve higher accuracy, while reducing the model size (floating-point operations) by up to 454.5x (8.9x). Therefore, the system is pervasive, efficient, yet very accurate.
• Can be used as a commercial application, for daily diagnosis and monitoring of diabetes, that’s easily implementable using a combination of
smart watches and smart phones
• Technology companies (e.g., Apple and Fitbit) may directly utilize the DiabDeep framework in their product
• Higher accuracy
• Reduces model size (floating-point operations) by up to 454.5x (8.9x)
• Efficient, pervasive, and very accurate
Stage of Development
The inventors have demonstrated the effectiveness of DiabDeep through a detailed analysis of data collected from 52 participants. For server (edge) side inference, they achieve a 96.3% (95.3%) accuracy in classifying diabetics against healthy individuals, and a 95.7% (94.6%) accuracy in distinguishing among type-1 diabetic, type-2 diabetic, and healthy individuals. Data collection is ongoing from more diabetes patients to further bolster the machine learning models (Diab-server and Diab-edge).
Hongxu Yin received the B.Eng. degree in Electrical and Electronics Engineering from Nanyang Technological University, Singapore, in 2015. He is currently a Ph.D. candidate in the Electrical Engineering Department at Princeton University. His research interests include execution-efficient deep neural networks, edge computing, and applied machine learning for smart healthcare and IoT applications.
Bilal Mukadam received his B.S.E in Electrical Engineering from Princeton University in 2019. His research interests include machine learning, wearable medical sensors, and the future intersection of healthcare and technology.
Xiaoliang Dai received the B.S. degree from Peking University, China, in 2014. He received his Ph.D. degree in Electrical Engineering from Princeton University in 2019. His research interests include automated neural network architecture synthesis, deep neural network compression, and efficient deep neural network adaptation for mobile devices.
Niraj K. Jha received his B.Tech. degree in Electronics and Electrical Communication Engineering from Indian Institute of Technology, Kharagpur, India in 1981, M.S. degree in Electrical Engineering from S.U.N.Y. at Stony Brook, NY in 1982, and Ph.D. degree in Electrical Engineering from University of Illinois, Urbana, IL in 1985. He is a Professor of Electrical Engineering at Princeton University. He is a Fellow of IEEE and ACM. He has served as the Editor-in-Chief of IEEE Transactions on VLSI Systems and an Associate Editor of IEEE Transactions on Circuits and Systems I and II, IEEE Transactions on Computer-Aided Design, IEEE Transactions on VLSI Systems, IEEE Transactions on Computers, IEEE Transactions on Multi-Scale Computing, Journal of Electronic Testing: Theory and Applications, and Journal of Nanotechnology. He is currently serving as an Associate Editor of Journal of Low Power Electronics. He has also served as the Program Chairman of the 1992 Workshop on Fault-Tolerant Parallel and Distributed Systems, the 2004 International Conference on Embedded and Ubiquitous Computing, and the 2010 International Conference on VLSI Design. He has served on more than 170 program committees. He has served as the Director of the Center for Embedded System-on-a-chip Design funded by New Jersey Commission on Science and Technology and as the Associate Director of the Andlinger Center for Energy and the Environment. He is the recipient of the AT&T Foundation Award and NEC Preceptorship Award for research excellence, NCR Award for teaching excellence, six Outstanding Teaching citations, and Princeton University Graduate Mentoring Award. He was given the Distinguished Alumnus Award by I.I.T., Kharagpur, in 2014. He has co-authored or co-edited five books. He has co-authored 15 book chapters and more than 440 technical papers. He has co-authored 14 papers that have won various awards and another six papers that have been nominated for best paper awards. He has received 17 U.S. patents. His research interests include machine learning, smart healthcare, cybersecurity, low power hardware/software design, computer-aided design of integrated circuits and systems, and monolithic 3D IC design. He has given several keynote speeches in the area of nanoelectronic design/test and smart healthcare.
Intellectual Property Status
Patent protection is pending.
Princeton University is currently looking for Industry collaborators to commercialize this technology.
Princeton University Office of Technology Licensing • (609) 258-7256• email@example.com