List your company

Ejection Fraction Estimation on EchoNet-Dynamic (test)

3.33MAEGT-mask EF proxy ceiling
2.756.66510.5814.495Aug 30, 2022Apr 19, 2023Dec 7, 2023Jul 26, 2024Mar 15, 2025Nov 2, 2025Jun 22, 2026

Evaluation results

MethodLinks
GT-mask EF proxy ceilingBackbone=N/A, Trainable components=none2026.063.33——4.33
EchoGNNNumber of parameters (x10^6)=1.72022.084.450.760.78—
BayesianNumber of parameters (x10^6)=31.52022.084.460.750.77—
Transformer (M)Number of parameters (x10^6)=346.8, Sampling Technique=M2022.085.950.520.55—
Direct EF regressorBackbone=CNN / video, Trainable components=full model2026.066.65——9.2
Transformer (R)Number of parameters (x10^6)=346.8, Sampling Technique=R2022.086.760.480.7—
EchoNet (AF)Number of parameters (x10^6)=31.52022.087.350.4——
BYOS + decoderBackbone=BYOS SSL, Trainable components=decoder2026.068.74——11.67
Partial FT BYOS + decoder (adapt + dec3)Backbone=BYOS SSL, Trainable components=decoder + sel. blocks2026.069.27——12.35
Partial FT DINOv3 + decoder (last 2 blocks + norm)Backbone=DINOv3 ViT, Trainable components=decoder + last 2 blocks + norm2026.069.57——12.62
DINOv3 + decoderBackbone=DINOv3 ViT, Trainable components=decoder2026.069.65——12.59
Partial FT BYOS + decoder (adapt only)Backbone=BYOS SSL, Trainable components=decoder + adapt2026.069.71——12.66
Supervised U-NetBackbone=CNN U-Net, Trainable components=full model2026.069.72——12.58
Partial FT DINOv3 + decoder (last block + norm)Backbone=DINOv3 ViT, Trainable components=decoder + last block + norm2026.0610.26——13.06
DINOv3 heuristicBackbone=DINOv3 ViT, Trainable components=none2026.0613.01——16.34
BYOS + linear probeBackbone=BYOS SSL, Trainable components=linear head2026.0613.49——17.05
DINOv3 + linear probeBackbone=DINOv3 ViT, Trainable components=linear head2026.0614.07——17.5
BYOS heuristicBackbone=BYOS SSL, Trainable components=none2026.0617.83——22.62