Multimodal deep learning enhances diagnostic precision in left ventricular hypertrophy

Abstract

Determining the etiology of left ventricular hypertrophy (LVH) can be challenging due to the similarity in (1) Aim: Determining the etiology of left ventricular hypertrophy (LVH) can be challenging due to the similarity in clinical presentation and cardiac morphological features of diverse causes of disease. In particular, distinguishing individuals with hypertrophic cardiomyopathy (HCM) from the much larger set of individuals with manifest or occult hypertension (HTN) is of major importance for family screening and the prevention of sudden death. We hypothesized that an artificial intelligence method based joint interpretation of 12 lead electrocardiograms and echocardiogram videos could augment physician interpretation. (2) Methods and Results: We chose not to train on proximate data labels such as physician over-reads of ECGs or echocardiograms but instead took advantage of electronic health record derived clinical blood pressure measurements and diagnostic consensus (often including molecular testing) among physicians in an HCM center of excellence. Using over 18,000 combined instances of electrocardiograms and echocardiograms from 2,728 patients, we developed LVH-Fusion. On held-out test data, LVH-Fusion achieved an F1-score of 0.71 in predicting HCM, and 0.96 in predicting HTN. In head-to-head comparison with human readers LVH-Fusion had higher sensitivity and specificity rates than its human counterparts. Finally, we use explainability techniques to investigate local and global features that positively and negatively impact LVH-Fusion prediction estimates providing confirmation from unsupervised analysis the diagnostic power of lateral T wave inversion on the ECG and proximal septal hypertrophy on the echocardiogram for HCM. (3) Conclusion: These results show that deep learning can provide effective physician augmentation in the face of a common diagnostic dilemma with far reaching implications for the prevention of sudden cardiac death.

High Throughput Deep Learning Detection of Mitral Regurgitation

“…28 This model architecture was previously used for other echocardiography tasks and shown to be effective. 17 The models were initialized with random weights and trained using a binary cross entropy loss function for up to 100 epochs, using an ADAM optimizer, an initial learning rate of 1e-2, and a batch size of 24 on two NVIDIA RTX 3090 GPUs. Early stopping was performed based on the validation loss.…”

“…Artificial intelligence (AI) has the ability to precisely phenotype subtle cardiac physiology as well as identify imaging features of disease severity not recognized by clinicians. [13][14][15][16] Deep learning has been applied to echocardiography to improve the precision of common measurements, such as left ventricular ejection fraction 13 and wall thickness 15,17 , as well as streamlining assessment of aortic stenosis 18 , hypertrophic cardiomyopathy (HCM) 15 , and cardiac amyloidosis (CA). 15,19 With increased ultrasound availability, AI guidance has been developed for both image acquisition and interpretation 13,20 .…”

Artificial Intelligence in Cardiology and Atherosclerosis in the Context of Precision Medicine: A Scoping Review

“…Madani et al [47] presented an algorithm to classify echocardiogram images with a 97.8% accuracy and no overfitting. No less important is the application of ML-based methods in the detection of wall motion abnormalities, assessment of the response of the cardiac muscle to the resynchronization therapy, prediction of major adverse cardiac events (MACEs) or coronary artery calcium (CAC), recognition, and assessment of valvular heart disease, classification of echocardiograms, differentiation of hypertrophic cardiomyopathy (HCM) and physiological hypertrophy of the athletes, or restrictive cardiomyopathy (RCM), and constrictive pericarditis [43,[48][49][50][51].…”

Artificial Intelligence in the Differential Diagnosis of Cardiomyopathy Phenotypes

“…The model was trained on more than 18,000 combined instances of examinations from 2728 patients. The proposed fusion model achieved an AUC of 0.92 (95% CI [0.862–0.965]), an F1-score of 0.73 (95% CI [0.585–0.842]), a sensitivity of 0.73 (95% CI [0.562–0.882]), and a specificity of 0.96 (95% CI [0.929–0.985]) [ 55 ].…”