Table 9 Performance comparison of the proposed framework with other studies.

From: A unified multi-task learning framework for automated assessment of left ventricular structure and its systolic function from echocardiography

Study

Primary Task

Dataset

Metric

Result

Sfakianakis et al27.

Segmentation

CAMUS

DSC

0.96

Amer et al30.

Segmentation

CAMUS

DSC

0.94

Wei et al32.

Segmentation

CAMUS

DSC

0.950

Ouyang et al.10

Segmentation

EchoNet-Dynamic

DSC

0.925

Chen et al44.

Segmentation

EchoNet-Dynamic

DSC

0.929

Duffy et al11.

Keypoint Detection

EchoNet-LVH

MAE

1.2–1.4 (mm)

Proposed MTL model

Segmentation

CAMUS

DSC

0.944

Segmentation

EchoNet-Dynamic

DSC

0.920

Keypoint Detection

EchoNet-LVH

MAE

1.13 (pixel)*

  1. * MAE for the proposed model is reported in pixels on the 112 × 112 standardized input image, representing the geometric localization error. This value is not directly comparable to metrics reported in millimeters (mm), which require patient-specific scaling factors.
Back to article page