School of Engineering Department of Computer Science and Engineering 84 AI + Healthcare: Research and Development of Intelligent Systems for Medical Diagnosis and Applications Supervisor: CHAN Gary Shueng Han / CSE Student: LIU Yuting / COGBM Course: UROP 1100, Spring Accurate fall detection is essential in smart home applications to prevent subsequent injuries and fatalities, especially among the elderly. Researchers have explored both vision-based and non-vision-based fall detection systems. Vision-based systems are popular due to their flexibility but face issues such as low accuracy, high computational costs, and privacy concerns. This study first examines the 60GHz Human Radar Detector, a non-vision-based system that relies on the Doppler Effect. Experimental results indicate limited coverage and inconsistent radar performance. To address these limitations, we developed a vision-based fall detection system using the YOLOv8 deep learning technique. Trained with the CAUCAFall Dataset, the YOLOv8 model demonstrated strong potential for integration into IoT-enabled smart homes, providing accurate and reliable fall detection. AI + Healthcare: Research and Development of Intelligent Systems for Medical Diagnosis and Applications Supervisor: CHAN Gary Shueng Han / CSE Student: WU Yongjin / COSC Course: UROP 3100, Fall CT Perfusion Parameter Prediction, an essential basis for the diagnosis and prognosis of Ischemic Stroke, is the task of calculating various perfusion parameters based on the brain CT scanning image. It was commonly performed by medical practitioners in the past, but the recent surge in Medical AI shows the potential of machine learning in handling this task. In this project, we will 1) introduce the task of CT Perfusion Parameter Prediction; 2) review several state-of-the-art deep generative models, including both unconditional and conditional; 3) discuss how generative models can be applied in the CT Perfusion Parameter Prediction and the potential challenge. AI Meets Big Data: User Analytics and Personalized Recommendation Based on Location Data Supervisor: CHAN Gary Shueng Han / CSE Student: REHMAN Abdul / COMP Course: UROP 2100, Fall The percentage of patients with hypertension and type 1 diabetes mellitus (T1DM) has been rising at an alarming rate, and for these individuals, an accurate prediction of blood glucose and pressure readings could mean the difference between life and death. Several deep learning models, including the powerful RNN architecture, have been tried and tested over the recent years to give precise predictions using past measurements data. However, many of these models ignore the impact of contextual data when making these predictions. In our proposed architecture, we combine the data-driven sequential component with a contextual layer to account for individual differences. In doing so, we separate the two components to allow a lightweight, fully personalized predictor which can be pretrained using metadata at user-end before being used for making predictions.
RkJQdWJsaXNoZXIy NDk5Njg=