School of Engineering Department of Computer Science and Engineering 94 AI + Healthcare: Research and Development of Intelligent Systems for Medical Diagnosis and Applications Recent breakthroughs in deep learning in the medical field have assisted practitioners in making more timely decisions than using manual image interpretation. However, a major issue has been the lack of data, due to problems such as ethical concerns. To mitigate the lack of data, transfer learning (TL) has been widely utilized, by taking parts of pre-trained models attacking other problems such as general image recognition. Successes have been reported in classification tasks such as medical image segmentation, but its efficacy has yet to be abundantly discussed in regression tasks such as CT perfusion mapping, which also suffer from the problem of limited data available due to high doses of radiation exposure in the scanning process. AI + Healthcare: Research and Development of Intelligent Systems for Medical Diagnosis and Applications Supervisor: CHAN, Gary Shueng Han / CSE Student: LI, Chun Chai Elton / COMP Course: UROP1100, Summer Detecting falls among elderly individuals is a critical aspect of ensuring their safety and well-being. The fall detection model serves as the backbone of the entire project, and this report presents an analysis of various existing action recognition models. This study adopts an approach to detect falls that involves utilizing skeleton-based action recognition model (ST-GCN, 2s-AGCN, PoseC3D, and STGCN++). The pipeline is built based on the Python framework provided by OpenMMLab, and the evaluation was done by using two fall datasets: UR Fall Detection Dataset and Le2i Fall Detection Dataset. These datasets contain videos of individuals performing daily activities in different environments, under camera positions. AI + Healthcare: Research and Development of Intelligent Systems for Medical Diagnosis and Applications Supervisor: CHAN, Gary Shueng Han / CSE Student: WU, Yongjin / COSC Course: UROP1100, Spring UROP2100, Summer Medical imaging is an important technique to facilitate diagnosis and prognosis in modern hospitals, and requires interdisciplinary knowledge including biology, physics, and computing technology. With the rapid advancement in machine learning and computer vision in recent decades, relevant techniques have also been proven useful in medical imaging. In our last UROP, we implement the model to detect brain tumor proposed by a published paper. In this project, we will focus on the issue of performance downgrade in small tumor exposed by previous experiment. Specifically, we study three questions: 1) whether such issue is general in medical imaging; 2) what methods have been proposed to solve this problem; 3) what additional contribution we could make to this field. Supervisor: CHAN, Gary Shueng Han / CSE Student: KUANG, Da / COGBM Course: UROP1000, Summer
RkJQdWJsaXNoZXIy NDk5Njg=