Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Assessing the Performance of the DINOv2 Self-supervised Learning Vision Transformer Model for the Segmentation of the Left Atrium from MRI Images

Bipasha Kundu, Bidur Khanal, Richard Simon, Cristian A. Linte

TL;DR

This work comprehensively evaluated the performance of DINOv2 for left atrial segmentation utilizing end-to-end fine-tuning and achieved, utilizing end-to-end finetuning, and achieved a mean Dice score of 87.1% and an Intersection over Union (IoU) of 79.2%.

Abstract

Accurate left atrium (LA) segmentation from pre-operative scans is crucial for diagnosing atrial fibrillation, treatment planning, and supporting surgical interventions. While deep learning models are key in medical image segmentation, they often require extensive manually annotated data. Foundation models trained on larger datasets have reduced this dependency, enhancing generalizability and robustness through transfer learning. We explore DINOv2, a self-supervised learning vision transformer trained on natural images, for LA segmentation using MRI. The challenges for LA's complex anatomy, thin boundaries, and limited annotated data make accurate segmentation difficult before & during the image-guided intervention. We demonstrate DINOv2's ability to provide accurate & consistent segmentation, achieving a mean Dice score of .871 & a Jaccard Index of .792 for end-to-end fine-tuning. Through few-shot learning across various data sizes & patient counts, DINOv2 consistently outperforms baseline models. These results suggest that DINOv2 effectively adapts to MRI with limited data, highlighting its potential as a competitive tool for segmentation & encouraging broader use in medical imaging.

Assessing the Performance of the DINOv2 Self-supervised Learning Vision Transformer Model for the Segmentation of the Left Atrium from MRI Images

TL;DR

This work comprehensively evaluated the performance of DINOv2 for left atrial segmentation utilizing end-to-end fine-tuning and achieved, utilizing end-to-end finetuning, and achieved a mean Dice score of 87.1% and an Intersection over Union (IoU) of 79.2%.

Abstract

Accurate left atrium (LA) segmentation from pre-operative scans is crucial for diagnosing atrial fibrillation, treatment planning, and supporting surgical interventions. While deep learning models are key in medical image segmentation, they often require extensive manually annotated data. Foundation models trained on larger datasets have reduced this dependency, enhancing generalizability and robustness through transfer learning. We explore DINOv2, a self-supervised learning vision transformer trained on natural images, for LA segmentation using MRI. The challenges for LA's complex anatomy, thin boundaries, and limited annotated data make accurate segmentation difficult before & during the image-guided intervention. We demonstrate DINOv2's ability to provide accurate & consistent segmentation, achieving a mean Dice score of .871 & a Jaccard Index of .792 for end-to-end fine-tuning. Through few-shot learning across various data sizes & patient counts, DINOv2 consistently outperforms baseline models. These results suggest that DINOv2 effectively adapts to MRI with limited data, highlighting its potential as a competitive tool for segmentation & encouraging broader use in medical imaging.

Paper Structure

This paper contains 9 sections, 1 equation, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Segmentation with pre-trained DINOv2
  4. Dataset Description
  5. Pre- and Post-processing
  6. Implementation Details
  7. Preliminary Results
  8. New Or Breakthrough Work To Be Presented
  9. Conclusion

Figures (3)

  • Figure 1: Detailed pipeline for fine-tuning DINOv2 on left atrium segmentation: utilizing labeled data to generate accurately predicted masks through transfer learning
  • Figure 2: Comparative analysis of data level few-shot learning performance across all methods: Left - evaluation of performance metrics with varying dataset sizes; Right - evaluation of performance metric with different patient counts
  • Figure 3: Qualitative comparison of binary segmentation results: Overlaying predictions (green) and ground truth (red) on input images for left atrium segmentation