JMIR Publications

ABSTRACT

Background:

In recent years, researchers have investigated machine learning (ML)-based approaches for the detection of left ventricular hypertrophy (LVH). However, the accuracy of ML in detecting LVH varies across different modeling variables and models. Systematic evidence is lacking in understanding how different ML approaches affect LVH detection accuracy.

Objective:

The objective of this study is to systematically assess their diagnostic accuracy to inform the development of artificial intelligence (AI) tools.

Methods:

PubMed, Embase, Cochrane Library, and Web of Science were comprehensively searched up to November 12, 2025. The Prediction model Risk Of Bias ASsessment Tool (PROBAST) was employed to evaluate the risk of bias. Subgroup analyses were performed based on ML model types and modeling variables (electrocardiogram (ECG), clinical features, and echocardiography (ECHO)). Only diagnostic 2×2 tables from validation sets were pooled for meta-analysis, with all statistical analyses performed using Stata.

Results:

A total of 25 articles were included in the analysis. The performance of ML models varied with input data types and algorithms. A meta-analysis showed that ECG-based models, in comparison, exhibited a SEN of 0.76 (95% CI: 0.66-0.84) and a SPE of 0.84 (95% CI: 0.78–0.89). ECHO-based models had a SEN ranging from 0.71 to 0.94 and a SPE ranging from 0.67 to 0.96. Models based on clinical features had a SEN of 0.78 (95% CI: 0.69–0.85) and a SPE of 0.71 (95% CI: 0.65–0.76). A subgroup analysis of ECG-based models revealed that the deep learning model produced a SEN of 0.71 (95% CI: 0.60–0.80) and a SPE of 0.79 (95% CI: 0.65–0.88).

Conclusions:

ML demonstrates reasonably high accuracy in detecting LVH. However, these conclusions are derived from limited evidence. Meanwhile, the extreme heterogeneity reported in the meta-analysis requires more critical interpretation. Current conclusions regarding model accuracy should be interpreted with caution. Therefore, future research should focus on constructing high-performance ML models based on imaging data for LVH diagnosis.