Development of a Deep Learning Workflow for Nuclear Segmentation and Classification in IHC-Stained Breast Cancer Slides with QuPath Integration

This project aims to develop an AI-driven pipeline for nuclear segmentation and classification in immunohistochemistry (IHC)-stained cancer slides, integrated into the open-source digital pathology platform QuPath. In collaboration with Alberta Health Services (AHS) and the University of Alberta, the project addresses a pressing need for scalable and interpretable tools to assist in cancer biomarker assessment—particularly where accurate detection of tumor nuclei is critical, such as in PD-L1 scoring for immunotherapy.

The intern will begin by curating a high-quality dataset of IHC whole slide images (WSIs), ensuring consistent annotations and staining protocols. Next, leading segmentation models like StarDist and PathoSAM will be benchmarked, followed by development of classifiers for nuclear subtypes using either multi-task learning or post-segmentation inference. The final pipeline will be deployed as a QuPath plugin using Python and PyTorch for seamless use in clinical workflows.

Deliverables include trained models, evaluation reports, visualization tools for quality control, and an operational QuPath plugin. External validation from a site in London, Ontario, will support generalizability. This project will enhance diagnostic accuracy and efficiency while contributing to Canada’s digital health infrastructure through open, reusable AI tools in pathology.

Faculty Supervisor:

Nilanjan Ray

Student:

Partner:

Alberta Health Services

Discipline:

Computer science

Sector:

Health and Related Sciences & Technology; Public administration; Retail trade

University:

University of Alberta

Program: