Bioengineering (Aug 2024)

Contrast-Enhancing Lesion Segmentation in Multiple Sclerosis: A Deep Learning Approach Validated in a Multicentric Cohort

  • Martina Greselin,
  • Po-Jui Lu,
  • Lester Melie-Garcia,
  • Mario Ocampo-Pineda,
  • Riccardo Galbusera,
  • Alessandro Cagol,
  • Matthias Weigel,
  • Nina de Oliveira Siebenborn,
  • Esther Ruberte,
  • Pascal Benkert,
  • Stefanie Müller,
  • Sebastian Finkener,
  • Jochen Vehoff,
  • Giulio Disanto,
  • Oliver Findling,
  • Andrew Chan,
  • Anke Salmen,
  • Caroline Pot,
  • Claire Bridel,
  • Chiara Zecca,
  • Tobias Derfuss,
  • Johanna M. Lieb,
  • Michael Diepers,
  • Franca Wagner,
  • Maria I. Vargas,
  • Renaud Du Pasquier,
  • Patrice H. Lalive,
  • Emanuele Pravatà,
  • Johannes Weber,
  • Claudio Gobbi,
  • David Leppert,
  • Olaf Chan-Hi Kim,
  • Philippe C. Cattin,
  • Robert Hoepner,
  • Patrick Roth,
  • Ludwig Kappos,
  • Jens Kuhle,
  • Cristina Granziera

DOI
https://doi.org/10.3390/bioengineering11080858
Journal volume & issue
Vol. 11, no. 8
p. 858

Abstract

Read online

The detection of contrast-enhancing lesions (CELs) is fundamental for the diagnosis and monitoring of patients with multiple sclerosis (MS). This task is time-consuming and suffers from high intra- and inter-rater variability in clinical practice. However, only a few studies proposed automatic approaches for CEL detection. This study aimed to develop a deep learning model that automatically detects and segments CELs in clinical Magnetic Resonance Imaging (MRI) scans. A 3D UNet-based network was trained with clinical MRI from the Swiss Multiple Sclerosis Cohort. The dataset comprised 372 scans from 280 MS patients: 162 showed at least one CEL, while 118 showed no CELs. The input dataset consisted of T1-weighted before and after gadolinium injection, and FLuid Attenuated Inversion Recovery images. The sampling strategy was based on a white matter lesion mask to confirm the existence of real contrast-enhancing lesions. To overcome the dataset imbalance, a weighted loss function was implemented. The Dice Score Coefficient and True Positive and False Positive Rates were 0.76, 0.93, and 0.02, respectively. Based on these results, the model developed in this study might well be considered for clinical decision support.

Keywords