Breast and fibroglandular tissue segmentation


Logo for Breast and fibroglandular tissue segmentation

About

Image Version:
1bfee743-6279-4d06-a080-2d7bf86252ee
Last updated:
May 6, 2022, 10:57 a.m.
Associated publication:
Samperna R, Moriakov N, Karssemeijer N, Teuwen J, Mann RM. Exploiting the Dixon Method for a Robust Breast and Fibro-Glandular Tissue Segmentation in Breast MRI. Diagnostics. 2022;12(7):1690.
Inputs:
  • T1 MRI  (Any T1 weighted MRI)
Outputs:
  • Breast and FGT Segmentation  (A breast and fibroglandular tissue segmentation map)

Model Facts

Summary

This algorithm produces a volumetric segmentation mask for the fatty tissue and the fibroglandular tissue (FGT) in the breast given as input either a fat (FS) or without fat suppression (WOFS) T1 breast MR acquisition. The output of this algorithm allows to easily calculate breast density from a FS or WOFS T1 breast MR acquisition.

Mechanism

The algorithm is a fully automated deep learning segmentation network. The network architecture is based on the nnUNet framework.

For more details, please refer to our article: Exploiting the Dixon Method for a Robust Breast and Fibro-Glandular Tissue Segmentation in Breast MRI

Validation and Performance

We compared a network trained with only without fat-suppression (WOFS-only) T1 breast MR acquisitions and a network trained with both WOFS and FS (WOFS + FS) T1 breast MRI acquisitions. Below you can find the Dice Similarity Coefficient (DSC) on 9 held-out test cases from our internal dataset for both networks.

Network trained on Test Breast DSC [95% CI] FGT DSC [95% CI]
WOFS-only WOFS 0.96 [0.94, 0.97] 0.92 [0.89, 0.95]
FS 0.10 [0.08, 0.12] 0.15 [0.11, 0.19]
WOFS + FS WOFS 0.96 [0.95, 0.97] 0.91 [0.87, 0.94]
FS 0.95 [0.94, 0.96] 0.86 [0.82, 0.91]

Uses and Directions

This algorithm was developed for research purposes only.

Warnings

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