A substantial demand exists to facilitate the machine learning (ML) models for on-site decision-making on intricate diseases and alert individuals about disease progression. Therefore, we conduct artificial intelligence (AI) research addressing critical health issues to catalyze breakthroughs in medical instrumentation. This project focuses on identifying non-invasive methods for diabetes diagnosis with biomedical signals and advanced deep-learning techniques. In addition, accurate blood glucose prediction, integration of predictive models into closed-loop insulin delivery, and predictive low glucose suspension (PLGS) are in active investigation. Furthermore, studies involve human activity recognition (HAR) and understanding the role of daily activities on blood glucose dynamics. By combining multi-sensory signals from wearable sensors, we develop software as a medical device (SaMD) and hardware medical devices (HMDs) for various health conditions, including diabetes, breathing disorders, fall detection, and diabetes-related cardiovascular and kidney diseases. This project seeks to advance the scientific foundation of AI in healthcare, fostering interdisciplinary collaboration.

Recent advances in machine learning (ML) algorithms have propelled artificial intelligence (AI) across diverse sectors, including healthcare, agriculture, and cyber-physical systems, enabling task automation, predictive capabilities, and learned decision-making. However, computation-intensive AI techniques pose accessibility challenges, hindering widespread deployment. Realizing ML models on edge devices for real-time applications faces computational, power, and memory constraints. Our research seeks solutions, including compact deep learning (DL) design, algorithm-hardware co-design, and efficient AI acceleration on edge devices while ensuring explainability, privacy, and security. In addition, the extension toward neuromorphic computing aims to explore energy-efficient and scalable AI implementations, potentially unlocking new avenues for edge computing applications. These will facilitate real-time inference and decision-making in various applications (e.g., autonomous vehicles, precision agriculture, robotics, healthcare diagnostics, and clinical decision support), thereby advancing the integration of AI with edge computing.

This project investigates the intersection of AI and semiconductor sectors, focusing on three major areas: sensor interface design, device modeling, and fabrication process optimization. Enhancing the efficiency and accuracy of sensor data acquisition is crucial for various applications such as IoT, medical devices, and environmental monitoring. Therefore, research will investigate CMOS integrated circuit design to advance sensing, localized computing, and interface electronics. In addition, we study machine learning applicability for integrated circuit design and semiconductor device research. For instance, incorporating machine learning for detecting fabrication defects is critical for ensuring high yields and reducing production costs. In addition, robust predictive models will facilitate improved performance and reliability in integrated circuits and semiconductor device fabrication.

Accurate and fast analysis of medical images is vital for effective diagnosis and treatment planning. This project focuses on computational biomedical image analysis, including confocal microscopy, ultrasound, computed tomography, etc., to enhance healthcare diagnostics, treatment, and patient care. The research harnesses cutting-edge machine learning and deep learning techniques with computer vision algorithms. For instance, our research includes developing various DL models for tasks such as microscopy image fusion, understanding microvascular changes in the disease process, blood vessel segmentation, etc., from different medical image modalities. Research also focuses on effectively acquiring, understanding, and analyzing bio-signals (e.g., ECG, EOG, EEG, PPG) to identify various diseases and health conditions. Subsequently, we validate these models and methods using real-world data and work on providing explainability for practical biomedical applications. We develop precise and efficient tools that contribute to improved healthcare outcomes, including early disease detection, personalized treatment strategies, and remote chronic condition management, thereby advancing the field of biomedical image and signal analytics in automated decisions.