Stage-wise mutual attention network for diagnostic support of chronic ischaemic heart disease using multimodal medical information

Background and Objective: Chronic ischaemic heart disease (CIHD) is the leading cause of death worldwide. Contributing factors include lifestyle related diseases such as diabetes and hypertension, as well as genetic predisposition. Diagnosing CIHD requires physicians to perform multiple tests, including highly invasive tests, which imposes a significant burden on patients. To address this issue, a diagnostic support network for CIHD has been developed, which uses multimodal medical information obtained solely from non-invasive tests as input. However, in this system, feature integration from different modalities is performed by simply concatenating each feature, which may not fully account for the diagnostic relevance of each input. This study aims to improve feature integration using an attention-based, stagewise fusion approach.Methods: We proposed a novel diagnostic support network for CIHD that employs an attention mechanism to integrate multimodal features. Our approach introduces a stage-wise integration approach where different feature vectors are combined progressively, two at a time. This allows features from different modalities to be gradually integrated while preserving diagnostically relevant information.Results: The proposed method demonstrated superior performance in distinguishing between healthy subjects and patients compared to conventional methods. Stepwise integration of three modalities, starting from closely related pairs, led to improved classification performance over approaches that merge all modalities simultaneously.Conclusions: Attention-based stage-wise feature integration provides effective support for CIHD diagnosis using non-invasive multimodal data. The gradual fusion strategy preserves diagnostic relevance and outperforms traditional concatenation methods, offering a promising direction for improving clinical decision support systems.