School of Engineering Department of Computer Science and Engineering 103 Machine Learning in Healthcare Supervisor: CHEN, Hao / CSE Student: KHAN, Asif / COMP LAI, Ka Wai / BIEN Course: UROP2100, Fall UROP1100, Fall The segmentation of medical images is an important topic due to the ever-increasing demand for fast automated diagnosis from medical images. As such, it is essential to develop and improve image segmentation algorithms. In this project, two methods, namely, MONAI and MedISeg, that are geared towards dealing with medical image segmentation extended on the basic UNET structure were investigated and compared using a few datasets. Comparing the two implementations, it was found that the MedISeg implementation has higher flexibility than the MONAI implementation as it provides more options for the user to specify. However, the MONAI implementation may appear to be more intuitive to readers. Multimodal Learning for Cancer Diagnosis and Prognosis Supervisor: CHEN, Hao / CSE Student: BHATT, Arun Datt / COMP Course: UROP1100, Fall Conventional deep learning approaches for cancer prognosis and diagnosis exploit the phenotypic properties within the tumor microenvironment of histopathological whole slide images (WSI) to predict the survival outcome of cancer victims. However, such an approach fails to consider the rich information in genomic and radiomic modalities. In this project, we collect multi-omics data for 14 cancer types from The Cancer Genome Atlas Program (TCGA) and compare co-attention mapping and fusion techniques to integrate the data for cancer prognosis. The report summarizes the learning objectives, background prerequisites, learning gains, and future direction of the project in reference to multimodal learning for cancer prognosis. Multimodal Learning for Cancer Diagnosis and Prognosis Supervisor: CHEN, Hao / CSE Student: CHOW, Chung Yan / SENG Course: UROP1000, Summer This course being the first UROP course being taken, much time was spent on literature review and implementation of the experiments. Unfortunately, due to some technical complications, only one study, Data Efficient and Weakly Supervised Computational Pathology on Whole Slide Images, was replicated.
RkJQdWJsaXNoZXIy NDk5Njg=