Physiology-Aware CNN and Zero-Shot Multimodal LLMs for ECG Image Classification: A Comparative Study

2026-06-22

AI Summary

Current zero-shot multimodal large language models (MLLMs) lack clinical reliability in classifying 12-lead electrocardiogram (ECG) images, struggling to effectively distinguish between normal and abnormal cases. This work proposes LeadGroupECG—a physiology-aware convolutional neural network that incorporates anatomical lead grouping priors—thereby embedding electrophysiological knowledge directly into the network architecture to optimize feature aggregation. Evaluated on both an internal test set and the external PTB-XL dataset, LeadGroupECG achieves ROC-AUC scores of 0.92–0.94 and 0.85–0.86, respectively, significantly outperforming baseline models such as ResNet18. In contrast, state-of-the-art MLLMs—including GPT-4.1, GPT-5.2, and Gemini-2.5 Pro—exhibit near-random performance (AUC ≈ 0.5), revealing fundamental limitations in zero-shot ECG interpretation.

Abstract

Multimodal large language models (LLMs) are increasingly adopted to interpret 12-lead ECG images, though the interpretations often lack validation. However, ECG image understanding significantly differs from general images as it depends on precise waveform morphology, lead relationships and accurate interval measurements. This study investigated whether zero-shot multimodal LLMs can reliably distinguish normal and abnormal ECG images and, in parallel, evaluated CNN-based models for clinically grounded references. Standard 12-lead ECG recordings were rendered as single-page images for a binary normal-abnormal classification task. Three prominent LLMs (GPT-5.2, GPT-4.1, and Gemini-2.5 Pro) were tested using a fixed zero-shot prompt across multiple runs. In parallel, a physiology-aware CNN-based model was developed with the capability to aggregate features from the predefined anatomical lead groups. The model was compared with ResNet18, DenseNet121, VGG16 baselines, and all the models were evaluated on an internal test set and external PTB-XL dataset. Across seeds, CNN-based models demonstrated stable discrimination, with average internal ROC-AUC of 0.92-0.94, and external ROC-AUC of 0.85-0.86. The proposed LeadGroupECG model significantly improved over its backbone internally without compromising external generalization. It remained competitive with other baselines, while consistently highlighting anatomical lead-group contributions. In contrast, zero-shot LLM discrimination remained near-chance (ROC-AUC around 0.5). The PR-AUC improved slightly when ECGs used a grid-based calibration background compared with the grid-free ECGs. Although multimodal LLMs can generate reasonable ECG narratives, their zero-shot diagnostic discrimination remains limited. Therefore, clinically framed, domain-specific architectures remain essential for AI-based ECG interpretation.

Problem

• ECG image classification
• zero-shot multimodal LLMs
• physiology-aware modeling
• 12-lead ECG
• clinical AI interpretation

Innovation

• Physiology-aware CNN
• Lead-group feature aggregation
• Zero-shot multimodal LLMs
• ECG image classification
• Anatomical interpretability