Skip to content
Back to portfolio

Applied ML Research

AnnotateAnyCell

Purdue research on efficient digital-pathology annotation combining segmentation, clustering, and interactive labeling.

Research Assistant at Purdue University · February 2024 – May 2025

Problem

Digital-pathology annotation is labor-intensive, creating a high burden for researchers labeling cell data from scratch.

My contribution

Purdue research project, advised by Prof. Ananth Grama, focused on efficient digital-pathology annotation using SAM- and Cellpose-based segmentation, clustering, semi-supervised learning, and interactive labeling, associated with the AnnotateAnyCell preprint.

Simplified architecture

Simplified: automated segmentation (SAM, Cellpose) and clustering feed a semi-supervised, interactive labeling workflow that reduces manual annotation effort.

Outcomes

  • Improved nuclei-labeling efficiency by approximately 30%.
  • Reduced labeled-data requirements by approximately 60%.
  • Reduced manual annotation effort by approximately 35%.

Technologies

SegmentationSAMCellposeClusteringSemi-supervised learningDeep learningInteractive labelingDigital pathology