Histopathology image segmentation

This project sits at the intersection of my two degrees. For the Computer Vision course, I built a UNet-based segmentation pipeline for identifying nuclei in histopathology images, followed by a nucleus classification system using both supervised learning and SimCLR-based contrastive learning.

The biology background mattered here in a way it doesn’t for most CV coursework. Understanding what the model is looking at — the morphological differences between cell types, the staining artefacts, the clinical significance of getting segmentation boundaries right — changes how you evaluate results. A Dice score is a number; knowing whether the errors are clinically meaningful is domain knowledge.