Abstract
Recent advancements in Large Language Models (LLMs) have drawn increasingattention since the learned embeddings pretrained on large-scale datasets haveshown powerful ability in various downstream applications. However, whether thelearned knowledge by LLMs can be transferred to clinical cardiology remainsunknown. In this work, we aim to bridge this gap by transferring the knowledgeof LLMs to clinical Electrocardiography (ECG). We propose an approach forcardiovascular disease diagnosis and automatic ECG diagnosis report generation.We also introduce an additional loss function by Optimal Transport (OT) toalign the distribution between ECG and language embedding. The learnedembeddings are evaluated on two downstream tasks: (1) automatic ECG diagnosisreport generation, and (2) zero-shot cardiovascular disease detection. Ourapproach is able to generate high-quality cardiac diagnosis reports and alsoachieves competitive zero-shot classification performance even compared withsupervised baselines, which proves the feasibility of transferring knowledgefrom LLMs to the cardiac domain.