HER2 biopsy analysis

The challenge

Analyzing breast cancer biopsies for the HER2 marker requires counting and separating cells in very high-resolution medical images, and HER2-low cases are especially hard to classify.

The solution

I evolved a local monolithic prototype into an interactive, decoupled web system (FastAPI + Streamlit) ready for Docker containers:

• Watershed segmentation for cell counting and separation, measured in square micrometers (µm²).
• Uncertainty index focused on HER2-low cases, with human-in-the-loop assistance: the system flags uncertain cases for professional review.
• Tile-based processing of high-resolution medical images (WSI).
• Clinical interoperability under the HL7 / FHIR standard.

My role

I handled the technical solution: image processing with OpenCV and Scikit-Image, the FastAPI API, the Streamlit interface and Docker packaging.

Outcome

The prototype went from running on a single local machine to a decoupled web application, with a published demo ready to grow toward a clinical environment.