Explainable AI-Enabled Wearable Sensors for Real-Time Thrombotic Event Early Warning in Connected Health Environments
- 13 March 2026
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (2) , 4549-4559
Abstract
Thrombotic events have multiple complications that remain difficult to identify early unless continuous non-invasive monitoring solutions are available to consumers; however, developing such products has been very restrictive thus far. In this article we provide an overview about a novel Explainable Artificial Intelligence (XAI) framework for wearable sensors and the earliest warning systems for thrombotic events in a connect health environment. Our framework is dedicated to supporting integration with mainstream consumer wearable technologies (smartwatches, fitness trackers) and allows for ease of implementation in existing consumer electronics ecosystems whilst still achieving clinical-grade prediction accuracy. We developed a method that integrates multimodal physiological signals from ECG-derived heart rate variability (HRV) as well as dynamics of skin temperature into a hierarchical attention based model that provides risk assessments that are interpretable and clinically relevant; Our extensive experiments show that the predictive accuracy of ThrombXAI is at 91.3% using an AUC value of 0.936 and represents improvements of 4.2% and 2.8% respectively to the current best available state-of-the-art methods for thrombotic event prediction. We further validated our results in a prospective study on 156 participants and compared the results to traditional screening approaches, in which we discovered statistically significant improvements; 89.3% sensitivity and 91.5% specificity for the early identification of thrombotic events. We provide the way forward with new paradigms of personalized preventative cardiovascular medicine using consumer wearables integrated with clinically actionable XAI technology1.Keywords
This publication has 35 references indexed in Scilit:
- MIMIC-IV, a freely accessible electronic health record datasetScientific Data, 2023
- InceptionTime: Finding AlexNet for time series classificationData Mining and Knowledge Discovery, 2020
- Investigating sources of inaccuracy in wearable optical heart rate sensorsnpj Digital Medicine, 2020
- Large-Scale Assessment of a Smartwatch to Identify Atrial FibrillationNew England Journal of Medicine, 2019
- Machine Learning in MedicineNew England Journal of Medicine, 2019
- High-performance medicine: the convergence of human and artificial intelligenceNature Medicine, 2019
- Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural networkNature Medicine, 2019
- Heart rate variability analysis in patients with multiple sclerosisMultiple Sclerosis and Related Disorders, 2018
- Passive Detection of Atrial Fibrillation Using a Commercially Available SmartwatchJAMA Cardiology, 2018
- Thrombophilia Testing and Venous ThrombosisNew England Journal of Medicine, 2017