Efficient screening for severe aortic valve stenosis using understandable artificial intelligence: a prospective diagnostic accuracy study

Abstract

Aims The medical need for screening of aortic valve stenosis (AS), which leads to timely and appropriate medical intervention, is rapidly increasing because of the high prevalence of AS in elderly population. This study aimed to establish a screening method using understandable artificial intelligence (AI) to detect severe AS based on heart sounds and to package the built AI into a smartphone-application. Methods and Results In this diagnostic accuracy study, we developed multiple convolutional neural networks (CNNs) using a modified stratified 5-fold cross-validation to detect severe AS in electronic heart sound data recorded at three auscultation locations. Clinical validation was performed with the developed smartphone application in an independent cohort (model establishment: n = 556, clinical validation: n = 132). Our ensemble technique integrating the heart sounds from multiple auscultation locations increased the detection accuracy of CNN model by compensating detection errors. The established smartphone application achieved a sensitivity, specificity, accuracy, and F1 value of 97.6% (41/42), 94.4% (85/90), 95.7% (126/132), and 0.93, respectively, which were higher compared to the consensus of cardiologists (81.0%, 93.3%, 89.4%, and 0.829, respectively), implying a good utility for severe AS screening. The Grad-CAM++ demonstrated that the built AIs could focus on specific heart sounds to differentiate the severity of AS. Conclusions Our CNN model combining multiple auscultation locations and exported on smartphone application could efficiently identify severe AS based on heart sounds. The visual explanation of AI decisions for heart sounds was interpretable. These technologies may support medical training and remote consultations.

Deep Learning for Heart Sound Analysis: A Literature Review

“…Based on this model, a smartphone application was established, achieving a 95.7% accuracy and 0.93 F1 score. Additionally, they employed Gradient-based Class Activation Maps to identify the specific heart sound features that the DL model focused on when distinguishing the severity of AS [62].…”

“…At last, 71 original articles were included. These studies can be broadly categorized into several groups: methods (15 papers, including heart sound segmentation [6, 7, 8, 9, 10, 11, 12, 13], noise cancellation [14, 15, 16], algorithm development [17, 18, 19], and database development [20]), cardiac murmurs detection (36 papers [21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56]), valvular heart disease (6 papers [57, 58, 59, 60, 61, 62]), congenital heart disease (4 papers [63, 64, 65, 66]), heart failure (4 papers [67, 68, 69, 70]), coronary artery disease (2 papers [71, 72]), rheumatic heart disease (2 papers [73, 74]), and extracardiac applications (2 papers [75, 76]).…”

“…In a study conducted by Luo et al using their smartphone application, over 80% of users were able to obtain good-quality heart sound recordings, with success rates independent of age, gender, body mass index, and smartphone versions [129]. As mentioned above, Makimoto et al have developed a smartphone application using a lightweight CNN model to detect severe AS [62]. Smartphone-based diagnostic models offer a convenient way for patients to monitor their health status at home, enabling the early detection of potential diseases and ensuring prompt access to treatment.…”

Diagnostic AI and Cardiac Diseases

“…Comparing each result in order to find the best ML system for the task, they later exported the system to a smart phone as an application where it achieved a 97.6% sensitivity, 94.4% specificity, 95.7% accuracy, and F1 value of 0.93. Compared with the consensus of cardiologists, these results were 81.0%, 93.3%, 89.4% and 0.829, respectively [ 24 ].…”

Algorithm for predicting valvular heart disease from heart sounds in an unselected cohort

“…Of the literature we identified, our study presented the highest accuracy for the prediction of AS. Makimoto et al predicted severe AS with an AUC of 0.968 when training directly on AS using convolutional neural networks and three auscultation positions ( 30 ). The study by Chorba et al ( 29 ) offered an interesting comparison, as their methodology is very similar to ours.…”