Evaluating self-supervised echocardiographic representations across downstream extraction strategies for left-ventricular segmentation and ejection fraction estimation

2026-06-22

AI Summary

This work addresses the limitation of existing self-supervised representation evaluation in medical imaging, which often relies on a single downstream strategy and fails to disentangle representation quality from downstream model capacity. To overcome this, the authors propose a multi-strategy evaluation framework that systematically compares heuristic feature extraction, frozen linear probing, lightweight decoder probing, and partial fine-tuning across left ventricular segmentation and ejection fraction estimation tasks. Experiments conducted with DINOv3 and BYOS self-supervised representations demonstrate that lightweight decoder probing achieves performance close to that of a supervised U-Net baseline, yielding Dice scores of 0.906 and 0.902 and ejection fraction mean absolute errors of 9.65 and 8.74, respectively. These findings underscore the decisive influence of downstream probing strategies on representation assessment and advocate for a refined evaluation paradigm in dense medical image analysis.

Abstract

Self-supervised learning (SSL) is increasingly used in medical imaging to reduce annotation requirements, but representation quality is often judged using a single downstream evaluation setting. For dense clinical tasks, this can confound representation quality with the capacity of the downstream model used to recover task-relevant information. We present a systematic evaluation of self-supervised representations for left-ventricular segmentation and ejection fraction (EF) estimation from apical four-chamber echocardiography on EchoNet-Dynamic. Rather than relying on a single downstream probe, we compare a hierarchy of extraction strategies with increasing expressivity: heuristic extraction without mask-supervised training, frozen linear probes, frozen lightweight decoder probes, and partial fine-tuning. We apply this framework to two complementary representation families: generic frozen self-DIstillation with NO labels (DINOv3) features and a task-adapted dense self-supervised representation, Bootstrap Your Own Segmentation (BYOS). In both families, heuristic extraction substantially understated what was recoverable from the frozen representation. For DINOv3, performance improved from Dice 0.684 and EF mean absolute error (MAE) 13.01 under heuristic extraction to Dice 0.906 and EF MAE 9.65 with a frozen lightweight decoder, approaching a supervised U-Net baseline (Dice 0.915, EF MAE 9.72). For BYOS, performance improved from Dice 0.687 and EF MAE 17.83 under heuristic extraction to Dice 0.902 and EF MAE 8.74 with a frozen lightweight decoder. These results show that conclusions about self-supervised representation quality in dense echocardiographic analysis depend strongly on the downstream extraction strategy used for evaluation. We therefore argue that multi-strategy evaluation is an important methodological consideration for SSL in dense medical image analysis.

Problem

• self-supervised learning
• echocardiography
• left-ventricular segmentation
• ejection fraction estimation
• downstream evaluation

Innovation

• self-supervised learning
• downstream evaluation
• echocardiography
• representation quality
• dense prediction