Medical University of Innsbruck (Dlaska Lab)
Model weights not public. Contact creators for more information.
Framework for training an echocardiography left-ventricle segmentation model purely by data-free knowledge distillation: a ConvLSTM-based student network learns to reproduce the masks produced by an EchoNet-Dynamic (DeepLabV3-ResNet50) teacher on entirely synthetic echo videos, with no real labeled data or even real videos required. Achieves state-of-the-art results identifying end-diastolic/end-systolic frames, reaching segmentation quality close to real-data training with substantially fewer weights; also introduces a human-annotation-free evaluation method using a large auxiliary model.
Architecture
RNN / LSTM / GRU
ConvLSTM-based video segmentation student network trained via data-free knowledge distillation from an EchoNet-Dynamic DeepLabV3-ResNet50 teacher on synthetic echo videos (EchoNet-Synthetic)
Framework
PyTorch
Added to catalog
2026-08-10
10,030 deidentified apical-4-chamber echo videos from Stanford Health Care. Source reports age and sex breakdowns.
Synthetic apical-4-chamber echocardiography videos generated by a diffusion model (Reynaud et al.), used to train the EchoDFKD student network without any real labeled data or real videos.
Left ventricle segmentation and end-diastolic/end-systolic frame identification from A4C echo video, trained via data-free knowledge distillation on synthetic video
Left ventricular ejection fraction, computed from predicted segmentation masks