Prostate segmentation by Guerbet

Logo for Prostate segmentation by Guerbet

About

Editors:
User Mugshot joeran.bosma  User Mugshot cabinader  User Mugshot noelie.debs  User Mugshot alexandre_bone  User Mugshot aroutier 
Contact email:
noelie.debs@guerbet.com
Image Version:
52cd84a8-0b9f-4bfa-84e6-2212c5275de8 — March 9, 2023

Summary

This model is based on nnU-Net for segmenting the prostate gland. The algorithm was trained on the PI-CAI dataset of 1500 patients, on a 5-fold cross-validation manner. It was trained by Guerbet IA team as part of their participation in the PI-CAI challenge.

Mechanism

Input:

  • Data type: axial T2w MRI, DWI with high b-value MRI, and ADC map.

  • File format: NIfTI or MHA format, containing pixel spacing information.

  • Target population: Men without prostactomy or resection.

Output: - Binary mask of the segmented prostate gland.

Model type: Baseline nnU-Net was trained with 5-fold cross-validation. Fold splits were given by the PI-CAI challenge.

Interfaces

This algorithm implements all of the following input-output combinations:

Inputs Outputs
1
    Transverse T2 Prostate MRI
    Transverse HBV Prostate MRI
    Transverse ADC Prostate MRI
    Transverse Prostate Segmentation MRI

Validation and Performance

On cross-validation dataset, nnU-Net performing prostate gland segmentation showed a mean overall 3D patient level dice of:

Fold 3D Dice
Fold 0 0.968
Fold 1 0.968
Fold 2 0.972
Fold 3 0.966
Fold 4 0.970

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