A Novel Continuous Left Ventricular Diastolic Function Score Using Machine Learning.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Left ventricular (LV) ejection fraction is a standard measure of systolic function, but a similar single metric for diastolic function is lacking.
• Assessing LV diastolic function traditionally relies on complex algorithms and multiple parameters.

Purpose of the Study:

• To develop a machine learning model for generating a continuous score to grade LV diastolic function.
• To utilize echocardiographic data for automated diastolic function assessment.

Main Methods:

• Trained machine learning models (SVM, DT, XGB, DNN) on 23,188 echocardiographic studies.
• Retrained a deep neural network (DNN) to predict a continuous LV diastolic function score (R-DNN).
• Validated the model on 5,798 studies using 2016 American Society of Echocardiography (ASE)/European Association of Cardiovascular Imaging (EACVI) guidelines.

Main Results:

• Machine learning models demonstrated high agreement with ASE/EACVI diastolic function grading (83%-100%).
• The continuous score correlated well with diastolic dysfunction severity: normal (1.00 ± 0.01), mild (0.74 ± 0.05), moderate (0.51 ± 0.06), severe (0.27 ± 0.11).
• The score effectively predicted abnormal diastolic function (AUC=0.99) and elevated filling pressures (AUC=0.99).

Conclusions:

• Machine learning can effectively process echocardiographic data to create an automated, continuous score for LV diastolic function.
• This novel score aligns well with current clinical guidelines for grading diastolic dysfunction.
• The automated score offers a promising tool for more precise and accessible diastolic function assessment.