From whole slides to individual cells
At the Bioinformatics Institute, A*STAR, I built an end-to-end pipeline for automated IHC biomarker expression analysis across whole-slide histopathology images.
The work combined tissue detection, stain normalization, watershed-based nuclei segmentation, and optical-density feature extraction. The pipeline produced 120-dimensional feature vectors for quantitative cell phenotyping.
Making analysis scalable
A Random Forest classification module achieved 97.9% accuracy in the project evaluation. Compared with the conventional IHC quantification workflow, the pipeline delivered approximately 10× higher throughput and reduced manual analysis effort by approximately 95%.
The modular architecture scaled to more than one million cells per slide, supporting reproducible quantification for downstream research.
These figures describe this research project’s evaluation, rather than a general clinical performance claim. The animated cell field on this website is a visual interpretation; it does not show patient data or microscopy results.