Decoding cerebral vasoregulation: an attention-based BiLSTM framework for elderly ischemic stroke prediction and biomarker identification with clinical associations

Stroke remains a major cause of mortality and long-term disability among older adults, with risk further increased by age-related decline in cerebral blood-flow regulation and vascular responsiveness. Traditional risk scores often overlook physiological dynamics, necessitating advanced multimodal analysis for early detection and prevention. This study hypothesizes that an attention-based bidirectional long short-term memory (BiLSTM) model can accurately classify stroke versus the control group using physiological biosignals while identifying interpretable individual biomarkers associated with cardiovascular risk indicators.

In this study, we have used the Cerebral Vasoregulation in Elderly with Stroke (CVES) dataset, comprising multimodal physiological biosignals data (e.g., ECG, arterial blood pressure, cerebral flow velocity, end-tidal \(CO_2\) etc) from 64 elderly participants. The proposed attention-based BiLSTM architecture encodes bidirectional temporal dependencies, applies self-attention for salient feature weighting, followed by dense layers for classification. Model evaluation has employed 5-fold and leave-one-out cross-validation, with SHapley Additive exPlanations (SHAP) for identification of individual biomarkers.

The proposed model also identifies an association between physiological and clinical data. We have achieved 98.6% test accuracy, with 0.996 AUC-ROC, 98.8% precision, and 98.9% recall, outperforming baselines like LSTM, GRU, and 1D-CNN. SHAP analysis highlights cardiovascular dominance (e.g., ECG, ABP) as primary biomarkers, with radiative heat loss and middle cerebral artery flow as key discriminators.

Clinical analysis revealed age-associated amplification of ECG and CO\(_2\) sensitivities, smoking/alcohol suppression of radiative attributions, and diabetes intensification across signals. The proposed model is also able to predict cardiovascular risk indicators using physiological data with \((R^2) = (0.8467 \pm 0.03)\). The proposed framework advances proactive stroke-risk assessment by fusing predictive accuracy with interpretable, personalized insights, surpassing literature benchmarks in multimodal integration and translational potential for elderly care.

We acknowledge the generous support of the Core Funds and the computing facilities provided by the Indian Institute of Information Technology Kota (IIIT Kota), which enabled the successful completion of this study. Department of Computer Science and Engineering, Sardar Vallabhbhai National Institute of Technology, Surat, Gujarat,, India Amity Institute of Biotechnology, Amity University, Noida, India Maitry Amitkumar Chauhan & Abhinandan Kaistha Department of Computer Science and Engineering, National Institute of Technology, Warangal, Telangana, India The authors declare no competing interests. All procedures were performed in compliance with relevant laws and institutional guidelines.

All data were anonymized and do not require further ethical approval. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material.