Projects

Deep learning-based breast cancer metastasis classification on whole slide images

The small size of tumor lesions poses a challenge for current Multiple Instance Learning (MIL) models in accurately classifying Whole Slide Images (WSI). To address this, we have introduced innovative weakly supervised learning techniques that effectively emphasize small tumor regions, resulting in enhanced accuracy and interpretability.

Deep learning to predict breast cancer recurrence risk from digital pathology images

Oncotype DX (ODX) is an accurate gene-assay for risk stratification of breast cancer patients and personalized therapy. However, ODX and similar gene assays are expensive, time-consuming, and tissue destructive. In this study, we predict ODX outcome based on readily available H&E stained histopathology images using our novel deep learning model.

Segment Anything Model for tumor bud segmentation

SAM establishes a groundbreaking benchmark for segmentation foundation models. Can we adapt it to cellular objects segmentation? My mentee, Sony, and I shown its feasibility in a colorectal tumor bud segmentation study.

Vision-language modeling for pathology images

Project is undergoing. We are developing generative model empowered by LLM and image models. Details will be exposed soon.