Medical devices, data, and AI for biomedical discovery and next-generation healthcare.
1. AI-Enabled Medical Devices
Access to medical services remains largely centered on hospitals, which can delay diagnosis until disease has reached an advanced stage. To bridge this gap, we develop novel home-based, wearable, and implantable electronic devices that continuously capture health data previously inaccessible in everyday settings. By combining these devices with AI models, we translate real-world data into clinically actionable insights. Examples include a diagnostic pen that captures handwriting biomechanics data for at-home Parkinson’s disease assessment; electronic textiles that stream physiological data for AI-based health monitoring; and a smart stent that captures postoperative blood-flow data for AI-enabled restenosis detection.
2. AI-Enabled Cardiovascular Care
Current gold-standard methods for diagnosing postoperative vascular stenosis, such as angiography and MRI, are centralized and episodic, often identifying disease only after complications arise. To address this limitation, we develop smart vascular grafts and stents that restore blood flow while continuously capturing cardiovascular biomechanical data. These devices have been validated in large-animal studies. AI-assisted analysis identifies stenosis-associated changes in these data to support timely diagnosis, while medical robotics delivers precise intervention. By combining previously inaccessible data with AI methods, we aim to shift the paradigm of cardiovascular care from reactive treatment to proactive management and improve outcomes for millions of patients.
3. Data-Driven Multiscale Mechanomedicine
Biomechanics plays a fundamental role in health across scales, from body movement and fine motor control to blood flow, cardiac motion, tissue mechanics, and cellular stiffness. Changes in these mechanical signatures can reveal disease onset and progression, treatment response, and recovery. We develop biomedical devices that capture previously inaccessible biomechanical data across these scales, alongside AI models that interpret the data. Together, these tools advance fundamental biomedical discovery and support translational applications such as biomechanics-based cancer screening.
4. Multimodal AI for Health
Interpreting heterogeneous, continuous health data and generating actionable insights in real time remain critical challenges. We develop multimodal AI methods that learn relationships across data streams from our medical devices while incorporating individualized clinical and contextual information. Reinforcement learning from clinician feedback can refine model outputs to improve interpretability, trustworthiness, and robustness. Our goal is to identify clinically meaningful patterns that support timely disease detection and dynamic risk alerts, enabling adaptive interventions to improve health outcomes.