School of Engineering Department of Computer Science and Engineering 106 Deep Learning for Medical Image Analysis Supervisor: CHEN Hao / CSE Student: ZHOU Siyu / COMP Course: UROP 1100, Spring UROP 2100, Summer The integration of advanced computer vision and natural language processing technologies has significantly enhanced medical diagnostics. Vision-language models, such as CLIP, enable robust pairing of medical images with textual descriptions, facilitating automated diagnostic processes. Convolutional neural networks, like ResNet, offer powerful capabilities for medical image classification. By leveraging these technologies, this project aims to streamline medical image analysis and improve diagnostic efficiency through automated pipelines. Diffusion Models for Medical Imaging and Analysis Supervisor: CHEN Hao / CSE Student: WANG Zihui / COMP Course: UROP 1100, Spring During this academic term, I adjusted my research focus by the supervisor’s guidance, prioritizing foundational algorithm studies over direct project participation. My independent learning centered on exploring mainstream AI image generation techniques through the systematic implementation of two representative models: Generative Adversarial Networks (GANs) and Denoising Diffusion Probabilistic Models (DDPMs). This report mainly documents my technical exercises including algorithm reproduction, model validation, and comparative analysis of output quality across different architectures. In the next stage, I will focus on theoretical deepening through recent literature review, performance optimization experiments, and practical applications in medical imaging scenarios, aiming to combine my theoretical understanding with the actual project. Medical Agent Ecosystem for Scalable Task Routing and Orchestration Supervisor: CHEN Hao / CSE Student: ZENG Zihao / COMP Course: UROP 1100, Summer This report summarizes my research process on chest X-ray abnormality detection using the open-source multi-modal large language model (MLLM) InternVL3B. The report details the process, from environment setup, model loading, and data processing to future inference and validation, meanwhile demonstrates the technology obstacles encountered and their solutions. In the process of project, I have successfully established the InternVL inference workflow on local machine and on the HPC4 cluster, processed the data into appropriate type and conducted zero-shot testing. The next move is to conduct fine-tuning based on the data-set and using the methods like adjust the prompts to improve the performance of model on chest X-ray abnormality detection.
RkJQdWJsaXNoZXIy NDk5Njg=