Skip to main content

← Gallery

DEPICTER

Deep representation clustering for histology annotation, shown on a whole-slide image of the CAMELYON17 dataset.

Open in TissUUmaps 3

DEPICTER (Deep rEPresentatIon ClusTERing) is an interactive segmentation tool for histopathology annotation. It produces a patch-wise dense segmentation map of a whole-slide image, in three steps:

  1. A pretrained model computes embeddings from image patches.
  2. The user selects a few benign and cancerous patches in the multi-resolution image.
  3. Guided by the embeddings, the labels spread to the rest of the image, by seeded iterative clustering or by gating in the embedding space.

Open the demo on the CAMELYON17 dataset.

Publication: E. Chelebian, C. Avenel, F. Ciompi and C. Wählby. DEPICTER: Deep representation clustering for histology annotation. Computers in Biology and Medicine, 108026, 2024. doi:10.1016/j.compbiomed.2024.108026

Edit this project