Work

Clinically Interpretable Zero-Shot ECG Classification via Multimodal Learning and Expert-Aligned Descriptors

Year: 2025

Type: conference-paper

Abstract: The widespread adoption of artificial intelligence in clinical cardiology is hampered by a critical factor: the lack of transparency in automated electrocardiogram (ECG) interpretation systems.While deep learning models can accurately classify cardiac abnormalities, their "black-box" nature prevents clinicians from verifying the diagnostic reasoning, undermining clinical trust.To address this, we developed a multimodal diagnostic framework that emulates clinical reasoning by directly linking ECG signal features to their corresponding textual descriptions from clinical reports.Our system learns to recognize abnormalities like Left Bundle Branch Block (LBBB) not merely as a classification label, but by identifying and associating it with established diagnostic criteria, such as a 'widened QRS complex (> 120 ms)'.Trained on a large dataset of paired ECG signals and narrative reports, our model achieves diagnostic accuracies comparable to conventional supervised models without requiring explicit training on classification labels.By grounding its predictions in clinically relevant subfeatures, the system provides transparent, verifiable evidence for its conclusions.This approach represents a paradigm shift toward AI systems that augment clinical decision-making with intelligible, evidence-based insights, fostering greater trust and facilitating integration into the diagnostic workflow.

Source: Computing in cardiology

DOI: 10.22489/cinc.2025.379