XCardioNet: An Explainable Cross-Modal Attention Hybrid Deep Learning Framework for Multimodal Cardiovascular Disorder Diagnosis
International Journal of Drug Delivery Technology · 01.07.2026 · 16 (60s)
DOI: 10.25258/ijddt.16.60s.7 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 01.07.2026. Verfasst von G. Amalorpavam und 1 weiteren Autor:innen.
Abstract (Originalsprache)
The fast-paced rise in cardiovascular morbidity across the globe, therefore highlights the urgency for intelligent diagnostics systems application for early, accurate and clinically interpretable assessment of subclinical cardiac conditions. Although remarkable progress has been made using deep learning in computer-aided healthcare, none explored the integrated analysis of complementary information encoded by raw ECG signals and heterogeneous patient-specific clinical indicators while existing cardiovascular screening models almost solely focused on isolated unimodal information. Such a discretized learning framework typically struggles to achieve effective feature generalization, strong context-agnostic reasoning power, and clear decision making interpretability that is quite important in high-stakes pathology settings. In order to address these limitations, this work present XCardioNet: a next-generation explainable multimodal cardiovascular intelligence framework that jointly fuses ECG morphology learning using advanced representation methods with clinical risk-aware representation fusion for robust tri-class cardiac disorder identification. This architecture then consists of a dual-channel preprocessing engine that improves fidelity in the multimodal data: structured clinical records train a K-nearest neighbor imputer to fill in missing features alongside an Isolation Forest optimization step for outlier detection and feature normalization, while raw ECG recordings are optimized through a sequence of Butterworth spectral filtering, wavelet-domain denoising, adaptive R-peak identification and heartbeat-wise segmentation to ultimately preserve diagnostically salient waveform morphology. Then, a hierarchical classifier based on convolutional neural networks, Bi-directional long short-term memory and Transformer attention is used to learn localized beat characteristics, long-range temporal rhythm dependencies and global sequence semantics from segmented ECG patterns. A unified model of cardiovascular risk - Meanwhile, an attention-based clinical representation learner is able to extract underlying latent pathological signatures from high-dimensional clinical feature space. In contrast to traditional late fusion methods, the proposed framework embeds both modalities into a shared latent manifold and performs dynamic interdependency modelling through the multihead cross-modal attention fusion mechanism thereby supporting patient-to-signal and signal-to-patient information exchange for adaptive diagnosis. For the sake of clinical trustworthiness, further enhance the framework through SHapley Additive exPlanations-based feature attribution and attention heatmap visualization for interpretable reporting of modality contribution and decision relevance. Through extensive experimental validation, it is demonstrated that XCardioNet achieves a notable 99.14% diagnostic accuracy while greatly exceeding standalone ECG analyses, isolated clinical prediction and non-attentive multimodal baselines with substantially improved consistency in precision, recall and F1-score across all cardiovascular predicts. Additionally, a real-time intelligent graphical diagnostic prototype is created to confirm the translated feasibility of the proposed model in practical point-of-care settings. The presented findings establish that explainable cross-modal deep attention fusion can serve as a clinically scalable and computationally reliable paradigm for next-generation precision cardiovascular screening and early cardiac event prevention
Zitieren
APA
Amalorpavam, G., & Rajkumar, N. (2026). XCardioNet: An Explainable Cross-Modal Attention Hybrid Deep Learning Framework for Multimodal Cardiovascular Disorder Diagnosis. International Journal of Drug Delivery Technology, 16(60s). https://doi.org/10.25258/ijddt.16.60s.7
Vancouver
Amalorpavam G, Rajkumar N. XCardioNet: An Explainable Cross-Modal Attention Hybrid Deep Learning Framework for Multimodal Cardiovascular Disorder Diagnosis. International Journal of Drug Delivery Technology. 2026;16(60s). doi:10.25258/ijddt.16.60s.7
BibTeX
@article{Amalorpavam2026,
author = {Amalorpavam, G. and Rajkumar, N.},
title = {XCardioNet: An Explainable Cross-Modal Attention Hybrid Deep Learning Framework for Multimodal Cardiovascular Disorder Diagnosis},
journal = {International Journal of Drug Delivery Technology},
year = {2026},
volume = {16},
number = {60s},
doi = {10.25258/ijddt.16.60s.7}
}
EUVIMED