Diagnostic accuracy of artificial intelligence in detecting left ventricular hypertrophy by electrocardiograph: a systematic review and meta-analysis
Noppachai Siranart, Natee Deepan, Witina Techasatian, Somkiat Phutinart, Walit Sowalertrat, Ponthakorn Kaewkanha, Patavee Pajareya, Nithi Tokavanich, Narut Prasitlumkum, Ronpichai Chokesuwattanaskul
Abstract
Several studies suggested the utility of artificial intelligence (AI) in screening left ventricular hypertrophy (LVH). We hence conducted systematic review and meta-analysis comparing diagnostic accuracy of AI to Sokolow-Lyon's and Cornell's criteria. Our aim was to provide a comprehensive overview of the newly developed AI tools for diagnosing LVH. We searched MEDLINE, EMBASE, and Cochrane databases for relevant studies until May 2023. Included were observational studies evaluating AI's accuracy in LVH detection. The area under the receiver operating characteristic curves (ROC) and pooled sensitivities and specificities assessed AI's performance against standard criteria. A total of 66,479 participants, with and without LVH, were included. Use of AI was associated with improved diagnostic accuracy with summary ROC (SROC) of 0.87. Sokolow-Lyon's and Cornell's criteria had lower accuracy (0.68 and 0.60). AI had sensitivity and specificity of 69% and 87%. In comparison, Sokolow-Lyon's specificity was 92% with a sensitivity of 25%, while Cornell's specificity was 94% with a sensitivity of 19%. This indicating its superior diagnostic accuracy of AI based algorithm in LVH detection. Our study demonstrates that AI-based methods for diagnosing LVH exhibit higher diagnostic accuracy compared to conventional criteria, with notable increases in sensitivity. These findings contribute to the validation of AI as a promising tool for LVH detection.