Early Pre-Stroke Detection via Wearable IMU-Based Gait Variability and Postural Drift Analysis

A research team has published a proof-of-concept study demonstrating how a minimal wearable sensor and AI can identify individuals at elevated risk of stroke before a major cerebrovascular event occurs. The system, developed by Chanakan Chaipan and Aueaphum Aueawatthanaphisut, uses a single inertial measurement unit (IMU) worn on the sacral region (lower back) to capture subtle pelvic motion during walking and standing. The pelvis acts as a biomechanical proxy for overall motor control, allowing the algorithm to quantify micro-instabilities in gait variability and postural drift—early digital biomarkers of neurological adaptation to impending stroke.

Raw sensor data is processed through a sensor fusion pipeline to extract kinematic features, which are then fed into a machine learning classifier for risk stratification. The model showed progressive increases in pelvic angular variability from healthy controls to the pre-stroke group, and finally to post-stroke patients, achieving a macro-averaged area under the curve (AUC) of 0.785. This indicates a preliminary but meaningful ability to discriminate between risk categories. The approach is not intended for clinical diagnosis but as a scalable, continuous screening tool.

The core innovation is treating pre-stroke motor impairment as a detectable, quantifiable signal rather than a purely clinical observation. By focusing on the pelvis—the body's center of mass—the system captures global motor control degradation that often precedes a stroke by days or weeks. This wearable configuration is significantly simpler and more affordable than multi-sensor or camera-based gait labs, opening the door to community-level deployment for at-risk populations, such as older adults or those with hypertension.