Deep and transfer learning for prostate gland segmentation in IHC images to automatically compartmentalise biomarker quantification

  • Category: CMMI
  • Swimlane: 2020-2021
  • Column: Assigned
  • Position: 1
  • Assignee: Christine Decaestecker
  • Creator: Olivier Debeir
  • Started:
  • Created: 18/05/2020 10:39
  • Modified: 19/04/2021 15:53
  • Moved: 19/04/2021 15:53
Description

Compartmentalising the quantitative evaluation of IHC biomarkers in a specific histological structure, such as glandular epithelium in prostatic lesions, is often required in histopathology to provide more relevant and informative measurements for clinical research (Gevaert et al., Sci Rep. 2018 Sep 25;8(1):14326. doi: 10.1038/s41598-018-32711-9). For this purpose, pathologists have to annotate thousands of structures present in histological slide series, a long, tedious and potentially biased task that would greatly benefit from automation.

This project aims to adapt a deep convolutional neural network trained to segment glandular epithelium in colorectal tumours (Van Eycke et al., Med Image Anal. 2018 Oct;49:35-45. doi: 10.1016/j.media.2018.07.004) to segment the same structures in prostatic tumours, using transfer learning and an image database set up at DIAPath-CMMI in collaboration with Dr. T. Gevaert (KUL). Performance assessment should take into account annotation imperfections.

Prerequisite:

programming (Python), Machine Learning, Image analysis

Promoter:

Prof. Christine Decaestecker (cdecas@ulb.ac.be)

Contact:

Egor Zindy (egor.zindy@ulb.ac.be)

Sub-Tasks
Internal links
Comments