Multimodal AI for CAncer Risk Assessment (MAICARA)
Aarthi Koripelly, 2nd-Year, Computer Science
- Abstract
- A personalized breast cancer risk screening program adjusts the imaging modality and frequency of screening exams according to a woman's risk of developing breast cancer. This approach has the potential to reduce costs and false positives by reducing unnecessary exams and finding more cancers at an earlier stage when they are still curable. Deep learning is a class of artificial intelligence algorithms that progressively extracts higher-level representations from raw input. We are developing Multimodal AI for Cancer Risk Assessment (MAICARA), which combines imaging, clinical, and genomic data into a single deep learning model to predict breast cancer risk. One critical challenge to applying deep learning techniques in the biomedical domain is the relative scarcity and expense of labeled patient data. Recently, new self-supervised methods which rely on contrastive losses have been shown to generate representations that are as good for classification as those generated using purely supervised methods. The advantage of contrastive loss functions is that they are able to extract features independent of labels associated with the pretraining dataset. We have collected over 30,000 mammograms and breast Magnetic Resonance Imaging (MRI) exams from over 10,000 patients at the University of Chicago Medical Center. Building on our existing prototype, we adapt the contrastive self-learning technique for MAICARA. We collate unlabeled MRI data from public datasets, refine our existing image preprocessing pipeline, and use those data to pretrain a model using the Momentum Contrast framework. We then fine-tune the model with our breast MRI imaging data for the task of breast cancer risk prediction. Finally, we optimize the hyperparameters and evaluate the performance of the model.
- Presented by
- Aarthi Koripelly
- Research Mentors
- Anna Woodard, UChicago Department of Medicine, Data Science Institute
- Other Affiliations
- College Global Health Scholar, Quad Undergraduate Research Scholar
- Keywords
- Biological & Health Sciences, Computing Science, Statistics





























































![Poster: Directed Synthesis and Testing of [2]Rotaxane Slide-Ring Gels](https://www.virtualpostersession.org/s/4736bfff55faa9cbdf1454bcf18c3b38/a9cd3a12a46efce9b76e857af879ee8f.png?u=2022-05-06%2014%3a02%3a57.680598989%20%2b0000%20UTC)
















