Endometrial Carcinoma classification


Logo for Endometrial Carcinoma classification

About

Creators:

User Mugshot svermorg 

User Mugshot f.ciompi 

User Mugshot tgelton 

Version:
3149da11-aefc-48cc-a5c6-ca477a7ecfe7
Last updated:
Aug. 23, 2022, 1:35 p.m.
Inputs:
  • Generic Medical Image 
Outputs:
  • Endometrium Attention Heatmap  (A heat map to denote where the model looked at for the prediction)
  • Likelihood of Endometrium (Pre)malignancy  (The likelihood of endometrium (pre)malignancy. Ranges from 0.0 to 1.0)

Model Facts

Summary

Contact information

  • Name of the model: Classification of (pre)malignancies in endometrium pipelle biopsies
  • Last edited: 27-07-2022
  • Email address: thijs.gelton@radboudumc.nl
  • Link to article: WIP

Description

This model classifies whether (pre)malignant tissue is present in a WSI. It, therefore, only requires a WSI as input and will from this produce both a likelihood of the (pre)malignancy being present and a heatmap, which indicates where in the WSI the tissue should be located. The model is based on CLAM, but carefully tuned to best fit pipelle biopsies.

Mechanism

Input:

  • Datatype: WholeSlideImage
  • File format: Aperio (.svs), Hamamatsu (.vms, .vmu, .ndpi), Leica (.scn), MIRAX (.mrxs) and Ventana (.bif).
  • Target group: women

Output:

  • Results: Likelihood between 0 and 1 stating the possibility of (pre)malignant tissue being present in the endometrium.
  • Type of image output: Probability map with a heat map color-coding* that indicates which regions in the image were influential to the prediction of the model: green and yellow to red colors respectively indicate low and high probability regions

*Beware: the heatmap will always show what might be (pre)malignant, even though the likelihood is low.

Validation and Performance

Ground truth: Translated report diagnoses into categories by AIOS pathologist at the Radboudumc, Nijmegen

Validation set:

Description Size Source Data Performance Metric Performance
Endometrium Pipelle Biopsy WSIs 2911 Radboud UMC, Nijmegen Oct. 2013 - April 2021 AUC 0.855 (0.818 - 0.855)

Ground truth: The majority vote of the pathologists participating in the interobserver variability study.

Test set (results obtained using all 5 folds in an ensemble):

Description Size Source Data Performance Metric Performance
Endometrium Pipelle Biopsy WSIs 91 Radboud UMC, Nijmegen Oct. 2013 - April 2021 AUC 0.960 (0.924 - 0.986)

Uses and Directions

This algorithm was developed for research purposes only.

In exchange for using this algorithm freely, we require the user to agree with us using their sample for future research

Warnings

Use this algorithm at your own risk. Do not use it for diagnosis of patients as this is still in a development phase.

Common Error Messages

Information on this algorithm has been provided by the Algorithm Editors, following the Model Facts labels guidelines from Sendak, M.P., Gao, M., Brajer, N. et al. Presenting machine learning model information to clinical end users with model facts labels. npj Digit. Med. 3, 41 (2020). 10.1038/s41746-020-0253-3