Key result
Full fine-tuning of a phase-aware TimeSformer model achieved an ROC-AUC of 0.893 and F1-score of 0.783 for HFrEF classification, outperforming a 3D ResNet18 baseline and classifier-only tuning.
Why the study?
Adaptation of video foundation models to medical imaging, particularly purpose-built models for spatio-temporal echocardiography videos to diagnose HFrEF, remains limited.
Does full fine-tuning of a phase-aware TimeSformer video foundation model improve automated diagnosis of HFrEF from apical four-chamber echocardiography videos compared to classifier-only tuning and a 3D ResNet18 baseline?
Does full fine-tuning of a phase-aware TimeSformer video foundation model improve automated diagnosis of HFrEF from apical four-chamber echocardiography videos compared to classifier-only tuning and a 3D ResNet18 baseline?
Full domain-adaptive fine-tuning of a general-purpose video foundation model (TimeSformer) with cardiac phase-aware sampling outperforms a standard 3D CNN baseline in automatically detecting HFrEF from apical four-chamber echocardiograms.
Answers are grounded in this paper and cite their sources.
May support automated HFrEF detection from A4C videos; hypothesis-generating and requires prospective validation before clinical adoption.
The rapid advancement of foundation models has transformed computer vision, yet their adaptation to medical imaging, particularly purpose built foundation models for spatio temporal echocardiography videos, remains limited. In this work, we explore the feasibility of fine tuning a pretrained foundation model for video (TimeSformer) for automated diagnosis of reduced ejection fraction heart failure (HFrEF) from apical four chamber videos. The model is finetuned on the publicly available RVENet A4C dataset with a phase aware clip generation strategy that captures complementary systolic and diastolic dynamics. We employ divided space time attention, layer wise learning rate decay, and multi clip inference to enhance robustness under limited data. Experimental results show that full fine tuning of the proposed model outperforms classifier only tuning and a 3D CNN baseline. These findings highlight the potential of leveraging foundation video models for echocardiography based heart failure diagnosis and motivates further domain adaptation studies in medical imaging.
No takes yet. Share an insight, caveat, or question.
Bolhassani et al. (2026) studied Heart failure with reduced ejection fraction (HFrEF, LVEF ≤40%) (n=619). Phase-aware TimeSformer video foundation model with full fine-tuning vs. Phase-aware TimeSformer with classifier-head-only fine-tuning (NFT) and a 3D ResNet18 baseline was evaluated on Automated binary classification performance (ROC-AUC) for distinguishing HFrEF from healthy subjects. Full fine-tuning of a phase-aware TimeSformer model achieved an ROC-AUC of 0.893 and F1-score of 0.783 for HFrEF classification, outperforming a 3D ResNet18 baseline and classifier-only tuning.
Synapse has enriched 10 related papers. Consider them for comparative context: