School of Engineering Department of Electronic and Computer Engineering 140 Artificial Intelligence Methods for Medical Videos Supervisor: LI Xiaomeng / ECE Student: LI Sze Yuen / CPEG Course: UROP 1000, Summer Our recent studies have focused on computer vision, with a particular emphasis on medical image segmentation. To begin, we explored early foundational papers on computer vision, specifically looking into classic models such as Convolutional Neural Networks (CNNs) and U-Net. We gained practical experience by working on tasks such as Skin Lesion Classification and Spleen CT Segmentation. Through reading various research papers, we discovered the promising potential of semi-supervised learning in addressing the challenge of limited annotated data in medical image segmentation. This approach enables training with a smaller amount of labeled data, although its accuracy currently does not surpass traditional models. Our study particularly focused on two papers by Haonan Wang: "DHC: Dual-Debiased Heterogeneous Co-training Framework for Class-Imbalanced Semi-supervised Medical Image Segmentation" and "Towards a Generic Semi-Supervised Framework for Volumetric Medical Image Segmentation", which propose methods to address issues such as imbalanced class distribution. We aim to deepen our understanding of these techniques by reproducing the code implementations to identify better optimization strategies. Artificial Intelligence Methods for Medical Videos Supervisor: LI Xiaomeng / ECE Student: LIU Yunfei / CPEG MA Wanqin / ELEC Course: UROP 4100, Fall UROP 4100, Fall This Fall semester, we analyzed the robust methods for medical image analysis, supervised by Professor Xiaomeng LI. Exploring techniques such as data augmentation, transfer learning, and attention mechanisms, our study aimed to enhance model performance on challenging medical datasets, paving the way for more accurate and reliable diagnostic applications. LIU Yunfei focuses on Unsupervised Domain Adaptation with Contrastive Learning for OCT Segmentation; details are shown in Part 2. MA Wanqin focuses on the details of SUV Adapters for Medical Disease Diagnosis Models, which are shown in Part 3. Federated Learning with Medical Images Supervisor: LI Xiaomeng / ECE Student: JANG Jae Won / ELEC Course: UROP 1100, Fall Federated Learning is a method of training machine learning models in a decentralized manner. This type of machine learning approach has been used widely in various fields and industries, including healthcare as a means to preserve data privacy and sovereignty when training machine learning models. Prior to embarking in research on the applications of federated learning in medical images, I first explored the field of medical image analysis and computer vision before going into greater depth or into more specialized research topics. I explored medical image analysis through learning about a range of different machine learning models and training them on a range of different datasets through using the Tensorflow deep learning framework.
RkJQdWJsaXNoZXIy NDk5Njg=