AI-Driven Predictive Analytics for Early Detection of Peripheral Artery Disease: Integrating Clinical Data and Imaging Insights

Abstract

Peripheral artery disease (PAD) constitutes a major and often under-recognized manifestation of systemic atherosclerosis, with substantial morbidity and mortality implications. Early detection remains challenging, owing to heterogeneous symptomatology, limited screening practices and subtle imaging changes in the asymptomatic or early stages. This research explores an integrative framework of artificial intelligence (AI)-driven predictive analytics that combines longitudinal clinical data, demographic and laboratory risk markers, and advanced vascular imaging features to enable early identification of PAD and stratification of progression risk. Leveraging machine-learning and deep-learning architectures, the proposed system ingests structured electronic health-record data, non-invasive vascular imaging (such as computed tomography angiography, magnetic resonance angiography and pulse-volume waveform recordings), and selected functional biomarkers to build multivariate risk models and imaging-derived phenotypes. A hybrid modelling pipeline is described, consisting of (i) feature extraction and engineering from clinical/instrumentation data, (ii) convolutional and transformer-based neural networks for vascular image segmentation and perfusion prediction, (iii) ensemble learning for risk-score computation, and (iv) temporal outcome modelling for early prediction of disease progression. Results from simulated and retrospective cohorts suggest that integration of imaging insights with standard clinical predictors significantly improves detection sensitivity in early or preclinical PAD compared with ankle–brachial index alone, and enables more accurate risk stratification of patients for limb- and cardiovascular-event outcomes. Key challenges including data harmonisation across imaging modalities, model interpretability, deployment in heterogeneous healthcare settings, and ethical considerations around algorithmic bias are discussed. The research concludes by proposing a roadmap for clinical translation of AI-enabled PAD detection: validation in large multicentre datasets, incorporation of explainability layers, periodic retraining with real-world data, and prospective trials to assess impact on clinical decision-making and patient outcomes.