Generating Contrast Enhanced MRI with Diffusion Models
Adam Yala, Professor
Computational Precision Health
Closed. This professor is continuing with Fall 2023 apprentices on this project; no new apprentices needed for Spring 2024.
Contrast-Enhanced Breast MRI is the most sensitive imaging modality for understanding who has breast cancer, what is the extent of the disease and what treatments may be most appropriate. The effectiveness of this modality critically relies on the injection gadolinium contrast which “lights up” lesions on imaging; however, this contrast agent permanently deposits in patient brains. This project aims to predict the contrast view of Contrast Enhanced MR from the multiple non-contrast series. From a computational perspective, the task is to predict a volume (MR after contrast) from other volumes (MR series pre contrast). We will explore novel diffusion modeling approaches, which has recently led to state of the art image generation models like StableDiffusion, to generate high resolution post contrast images. Given a contrast prediction model, we are also interested in quantifying its uncertainty to identify subsets of the population where the model is sufficiently accurate to skip contrast.
Role: Students will work with UCSF MRI data for the above project and they will lead all model development and experimentation. Students will be trained to engage in the relevant research literature, formulate research problems, build effective and modular code-bases and to engage in scientific writing.
Qualifications: Applicants should have knowledge of Python, PyTorch and strong background/interest in machine learning and Computer Vision. Prior experience in building generative models is desirable but not essential. Students are expected to meet with the faculty mentor twice a week.
Hours: 12 or more hours
Biological & Health Sciences Digital Humanities and Data Science