Patient-Level Multimodal Question Answering from Multi-Site Auscultation Recordings
This work addresses the problem of subjective auscultation interpretation for medical diagnostics, offering a scalable method that is incremental by building on existing ALMs and LLMs.
The paper tackles the challenge of interpreting multi-site auscultation recordings for patient-level assessment by aligning them with a frozen LLM embedding space, achieving state-of-the-art results of 0.865 F1-macro and 0.952 BERTScore on the CaReSound benchmark.
Auscultation is a vital diagnostic tool, yet its utility is often limited by subjective interpretation. While general-purpose Audio-Language Models (ALMs) excel in general domains, they struggle with the nuances of physiological signals. We propose a framework that aligns multi-site auscultation recordings directly with a frozen Large Language Model (LLM) embedding space via gated cross-attention. By leveraging the LLM's latent world knowledge, our approach moves beyond isolated classification toward holistic, patient-level assessment. On the CaReSound benchmark, our model achieves a state-of-the-art 0.865 F1-macro and 0.952 BERTScore. We demonstrate that lightweight, domain-specific encoders rival large-scale ALMs and that multi-site aggregation provides spatial redundancy that mitigates temporal truncation. This alignment of medical acoustics with text foundations offers a scalable path for bridging signal processing and clinical assessment.