Predictive Spatio-Temporal Analytics for Early Detection of Lower Limb Ischemia Using Multi-Modal Clinical Data

Abstract

Lower limb ischemia remains one of the most dangerous vascular conditions due to its rapid progression and the narrow time window for clinical intervention. Traditional diagnostic pathways rely heavily on clinician judgment, intermittent imaging, and episodic physiological measurements, which often fail to capture the dynamic onset of ischemic deterioration. This study introduces a predictive spatio-temporal analytics framework that integrates multi-modal clinical data, including Doppler waveforms, perfusion indices, thermographic patterns, gait-cycle temporal signals, and vascular laboratory markers. Using a hybrid deep learning architecture combining convolutional feature extraction, temporal encoding, and probabilistic risk modelling, the system predicts ischemic onset up to 48 hours earlier than standard clinical assessment. Spatial tissue signatures extracted from thermography and duplex ultrasound were fused with sequential hemodynamic measurements using an attention-based encoder, enabling robust early-stage detection even in irregular, noisy clinical environments. Model interpretability methods were incorporated to ensure transparent decision reasoning, highlighting anatomical regions and temporal segments contributing to high-risk predictions. Results demonstrate substantial improvements in sensitivity, reduced false negatives, and enhanced early risk stratification for patients with peripheral arterial disease. The findings underscore the potential of spatio-temporal clinical analytics to transform ischemia screening, triage, and intervention planning in vascular care settings.