Related Experiment Video
Updated: Nov 5, 2025
06:34
Evaluation of Left Ventricular Structure and Function using 3D Echocardiography
Published on: October 28, 2020
4.2K
Automated Left Ventricular Dimension Assessment Using Artificial Intelligence Developed and Validated by a UK-Wide
James P Howard1,2, Catherine C Stowell1,2, Graham D Cole1,3
1Imperial College London (J.P.H., C.C.S., G.D.C., K.A., K.V., D.P.F., M.J.S.-S.).
Circulation. Cardiovascular Imaging
|May 17, 2021
Summary
A collaborative artificial intelligence (AI) trained on expert echocardiography data performed comparably to individual human experts in measurements. This approach establishes a new standard for AI development in medical imaging.
Area of Science:
- Cardiology
- Medical Imaging
- Artificial Intelligence
Background:
- Echocardiography analysis requires expert-level training and validation.
- Developing standardized AI techniques necessitates large, expert-annotated datasets.
- The Unity Collaborative was established to aggregate expertise from 17 hospitals.
Purpose of the Study:
- To develop an online platform for collaborative AI training in echocardiography.
- To create a robust dataset of expert annotations for AI model development.
- To standardize AI performance against human expert consensus.
Main Methods:
- Trained a convolutional neural network using 2056 frames from 1265 echocardiography video-loops annotated by 9 experts.
- Validated the AI using 100 new video-loops with 13 experts providing double annotations.
- Established a 26-opinion consensus as the reference standard for validation.
Main Results:
- AI precision for left ventricular internal dimension (3.5 mm) surpassed individual expert precision (4.4 mm).
- AI demonstrated superior performance in interventricular septum thickness and posterior wall thickness measurements compared to individual experts.
- High intraclass correlation coefficients (0.926 for AI vs. 0.817 for experts) indicate strong AI agreement with consensus.
Conclusions:
- Multi-institutional expert collaboration successfully developed a high-performing AI for echocardiography.
- The developed AI achieved performance levels comparable to individual human experts.
- Future echocardiographic AI research should adopt expert consensus as the reference standard.
