Artificial Intelligence-Assisted Left Ventricular Diastolic Function Assessment and Grading: Multiview Versus Single...

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Clinical assessment of left ventricular diastolic function (LVDF) is complex, requiring multiple echocardiographic parameters and experienced interpretation.
• Current LVDF assessment methods are time-consuming and rely heavily on clinician expertise.
• There is a need for efficient and accurate tools to aid in LVDF evaluation.

Purpose of the Study:

• To develop an artificial intelligence (AI)-assisted system for the clinical assessment and grading of LVDF.
• To improve the efficiency and accuracy of LVDF evaluation using AI.
• To reduce the reliance on expert interpretation and time-intensive processes.

Main Methods:

• Developed AI models using a large dataset (1,304 studies) for view classification and segmentation of echocardiographic images.
• Trained AI models on 2,150 studies with expert-defined LVDF labels for single-view classification (strain or video interpretation).
• Validated AI model performance on an external dataset of 388 prospective studies.

Main Results:

• AI view classification and segmentation models demonstrated high accuracy (sensitivity >0.9, IoU >0.8).
• AI quantification of 2D and Doppler images showed narrow limits of agreement with expert assessments.
• AI models accurately detected and graded LV diastolic dysfunction (DD) with accuracies of 0.9 and 0.92, respectively, in seconds.

Conclusions:

• AI models provide LVDF assessment and grading comparable to human experts following guideline-based algorithms.
• AI can assess LVDF using single-view 2D strain or video when Doppler data is unavailable.
• The developed AI system has the potential to significantly reduce labor, costs, and improve workflow efficiency for clinical LVDF assessment.