Artificial Intelligence in Detecting Cardiac Function

Overview

The Korotkoff Sounds（KS）, which have been in use for over a century, are widely regarded as the gold standard for measuring blood pressure. Furthermore, their potential extends beyond diagnosis and treatment of cardiovascular disease; however, research on the KS remains limited. Given the increasing incidence of heart failure (HF), there is a pressing need for a rapid and convenient prehospital screening method. In this study, we propose employing deep learning (DL) techniques to explore the feasibility of utilizing KS methodology in predicting functional changes in cardiac ejection fraction (LVEF) as an indicator of cardiac dysfunction.

Blood Pressure Measurement：Around 72 hours before and after the completion of the patient\'s echocardiogram, considering the variability in the patient\'s blood pressure and ejection fraction at different times, blood pressure should be measured in each participant at least twice a day, up to a maximum of six times. Each patient should be instructed to remain in a quiet state for 10 minutes before blood pressure measurement. Blood pressure measurement should be conducted according to the following criteria: the cuff used to measure blood pressure should be wrapped around the patient\'s arm above the elbow joint, positioned 2-3 cm above the level of the heart, with a snugness that allows one finger to fit underneath. Place the stethoscope head at the brachial artery pulse point on the left elbow joint, then begin inflation. Inflate continuously until the sound of the pulse beat disappears; then inflate an additional 20 mmHg before stopping inflation. Slowly deflate while listening-the first audible pulse beat is the systolic pressure, and the disappearance of the pulse sound is the diastolic pressure. Use the Hanhong POPULAR-3 electronic stethoscope to record the aforementioned process, with each audio recording lasting 25 seconds uniformly. Data Analysis Overview: In terms of data analysis, deep learning models are developed based on Torch version 1.5.0, utilizing Transformer network architecture to analyze the collected audio data. Network One: Identifying the presence of cardiac functional abnormalities through Korotkoff sounds. Evaluation Metrics: Receiver Operating Characteristic (AUROC), sensitivity, specificity, and F1 score (harmonic mean of sensitivity and specificity) to assess model performance on the test dataset. Network Two: NYHA classification of Korotkoff sounds. Evaluation Metrics: Confusion matrix, weighted accuracy, multi-class ROC curve, F1 score. Network Three: Heart failure classification of Korotkoff sounds in heart failure patients. Evaluation Metrics: Confusion matrix, weighted accuracy, multi-class ROC curve, F1 score. Network Four: Left Ventricular Ejection Fraction (LVEF) prediction from Korotkoff sounds. Evaluation Metrics: Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), R2 Score.