School of Engineering Department of Electronic and Computer Engineering 139 Deep Learning for Multi-class Retinal Disease Classification in Real-World Setting Supervisor: LI Xiaomeng / ECE Student: LIU Zichen / DSCT Course: UROP1100, Spring This semester, I participated in UROP under Professor LI, Xiaomeng's supervision. My work includes reading relevant papers, implementing and reappearing recent papers' experiment results, and conducting experiments on novel ideas. Specifically, I read more than 30 latest works about medical image and computer vision. I reappeared the result of the paper “A Fourier-based Framework for Domain Generalization” and “RSCFed: Random Sampling Consensus Federated Semisupervised Learning”. I also experimented to verify that MMD loss can improve the result of federated semi-supervised learning classification. In this final report, I will illustrate my works in detail. Deep Learning for Multi-class Retinal Disease Classification in Real-World Setting Supervisor: LI Xiaomeng / ECE Student: ZHANG Yuyao / DSCT Course: UROP2100, Summer I read some papers during this summer UROP program and participated in a challenge. Therefore, in the first part of the report, I will summarize some papers, and in the second part, I will discuss the challenge. Federated Learning with Medical Images Supervisor: LI Xiaomeng / ECE Student: MU Xihe / CPEG Course: UROP1000, Summer In this summer, we are assigned to the group led by mentor Yao Huifeng. Ma Wanqin focus on preparing Cardiac MRI Analysis Challenge under Respiratory Motion (CMRxMotion), MICCAI 2022. Mu Xihe is assigned to read papers about selftraining and consistency learning(eg: pseudo label and mean teacher). To research well, the main goal is to apply what we learned with Professor.Li Xiaomeng during past UROP. After careful selection, we finally decide to use method CACPS to train effective model and participate in the challenge. The method mainly focus on practicing effective segmentation on medical images with the use of pseudo labels. This report will provide challenge information and the details of experiments I did in summer.
RkJQdWJsaXNoZXIy NDk5Njg=