Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Total Variation Regularized Framework for Epilepsy-Related MRI Image Segmentation

Mehdi Rabiee, Sergio Greco, Reza Shahbazian, Irina Trubitsyna

TL;DR

This work tackles FCD segmentation in 3D brain MRI under limited annotated data. It extends a transformer-based MS-DSA-Net by incorporating an anisotropic Total Variation loss to enforce spatial smoothness directly during training, reducing reliance on post-processing. Compared to a baseline Dice (and Dice+ BCE) setup, the TV-regularized model achieves a Dice improvement of about 11.9% and a 13.3% gain in precision, while also reducing false positive clusters by roughly 61.6%. The approach enhances segmentation consistency and has potential utility for other small-lesion detection tasks in 3D medical imaging.

Abstract

Focal Cortical Dysplasia (FCD) is a primary cause of drug-resistant epilepsy and is difficult to detect in brain {magnetic resonance imaging} (MRI) due to the subtle and small-scale nature of its lesions. Accurate segmentation of FCD regions in 3D multimodal brain MRI images is essential for effective surgical planning and treatment. However, this task remains highly challenging due to the limited availability of annotated FCD datasets, the extremely small size and weak contrast of FCD lesions, the complexity of handling 3D multimodal inputs, and the need for output smoothness and anatomical consistency, which is often not addressed by standard voxel-wise loss functions. This paper presents a new framework for segmenting FCD regions in 3D brain MRI images. We adopt state-of-the-art transformer-enhanced encoder-decoder architecture and introduce a novel loss function combining Dice loss with an anisotropic {Total Variation} (TV) term. This integration encourages spatial smoothness and reduces false positive clusters without relying on post-processing. The framework is evaluated on a public FCD dataset with 85 epilepsy patients and demonstrates superior segmentation accuracy and consistency compared to standard loss formulations. The model with the proposed TV loss shows an 11.9\% improvement on the Dice coefficient and 13.3\% higher precision over the baseline model. Moreover, the number of false positive clusters is reduced by 61.6%

A Total Variation Regularized Framework for Epilepsy-Related MRI Image Segmentation

TL;DR

This work tackles FCD segmentation in 3D brain MRI under limited annotated data. It extends a transformer-based MS-DSA-Net by incorporating an anisotropic Total Variation loss to enforce spatial smoothness directly during training, reducing reliance on post-processing. Compared to a baseline Dice (and Dice+ BCE) setup, the TV-regularized model achieves a Dice improvement of about 11.9% and a 13.3% gain in precision, while also reducing false positive clusters by roughly 61.6%. The approach enhances segmentation consistency and has potential utility for other small-lesion detection tasks in 3D medical imaging.

Abstract

Focal Cortical Dysplasia (FCD) is a primary cause of drug-resistant epilepsy and is difficult to detect in brain {magnetic resonance imaging} (MRI) due to the subtle and small-scale nature of its lesions. Accurate segmentation of FCD regions in 3D multimodal brain MRI images is essential for effective surgical planning and treatment. However, this task remains highly challenging due to the limited availability of annotated FCD datasets, the extremely small size and weak contrast of FCD lesions, the complexity of handling 3D multimodal inputs, and the need for output smoothness and anatomical consistency, which is often not addressed by standard voxel-wise loss functions. This paper presents a new framework for segmenting FCD regions in 3D brain MRI images. We adopt state-of-the-art transformer-enhanced encoder-decoder architecture and introduce a novel loss function combining Dice loss with an anisotropic {Total Variation} (TV) term. This integration encourages spatial smoothness and reduces false positive clusters without relying on post-processing. The framework is evaluated on a public FCD dataset with 85 epilepsy patients and demonstrates superior segmentation accuracy and consistency compared to standard loss formulations. The model with the proposed TV loss shows an 11.9\% improvement on the Dice coefficient and 13.3\% higher precision over the baseline model. Moreover, the number of false positive clusters is reduced by 61.6%

Paper Structure

This paper contains 20 sections, 3 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Contributions.
  3. Organization.
  4. Related Works
  5. Proposed Model
  6. Loss Functions
  7. Dice Loss
  8. Binary Cross Entropy (BCE) Loss:
  9. Total Loss:
  10. Experiments
  11. Dataset and Preprocessing
  12. Training
  13. Evaluation Metrics
  14. Post-processing
  15. Quantitative Results
  16. ...and 5 more sections

Figures (3)

  • Figure 1: The architecture of the system for FCD detection. It is based on the MS-DSA-Net (inside the red box) and the proposed TV loss function.
  • Figure 2: Comparison of predicted segmentation before and after post-processing for the base model.
  • Figure 3: Segmentation results with the proposed TV loss, before and after post-processing.