Artificial Intelligence-Enhanced Electrocardiogram for the Early Detection of Cardiac Amyloidosis

Comprehensive Summary

Grogan et al. developed a deep neural network to detect cardiac amyloidosis (CA) using standard 12-lead electrocardiograms from 2,541 patients with CA (AL or ATTR) and 2,454 age- and sex-matched controls at Mayo Clinic. The ECG dataset was divided into training (60%), validation (20%), and testing (20%) cohorts, and the model analyzed raw ECG waveforms without handcrafted features. In the holdout test set, the AI-ECG achieved an AUC of 0.91, with sensitivity of 0.84 and specificity of 0.85, detecting 426 (84%) of patients with CA at the optimal probability threshold. Performance remained consistent across amyloid subtypes. Among patients with earlier ECGs available, the model predicted CA more than six months before clinical diagnosis in 59% of patients with prediagnosis ECGs. Simplified versions also performed well, including a single-lead model (best lead V5, AUC 0.86), and a 6-lead model (AUC 0.90). The authors concluded that AI-enhanced ECG analysis can accurately detect CA and may help enable earlier diagnosis of the life-threatening disease.

Outcomes and Implications

Cardiac amyloidosis is frequently underrecognized and often diagnosed late, despite therapies that improve outcomes when started earlier. This study suggests that AI applied to routine ECGs can detect subtle electrical patterns associated with amyloid infiltration that may not be apparent on conventional interpretation. Because ECG is inexpensive and widely accessible, integrating AI analysis into routine workflows could potentially help flag patients for further evaluation with imaging or specialty referral. The ability to deploy similar models on single-lead or 6-lead ECG devices also raises the possibility of point-of-care or wearable screening in at-risk populations. However, the model was developed using data from a single center, so external validation in broader populations will be essential before widespread clinical adoption.