Abstract Body: Smartphone-based pupillometry offers potential for widespread neurological monitoring in critical care, including for stroke patients in both hospital and post-hospital rehabilitation settings. However, varying measurement conditions, operator-dependent differences, and device models affect result reliability [1]. We developed an AI-powered app with an invariant Pupil Reactivity (PuRe) score to enable standardized quantitative pupillometry across diverse settings, addressing the need for consistent neurological monitoring by different healthcare providers under varying conditions [2, 3].
We conducted two studies to validate PuRe score stability. At Nicolaus Copernicus University (Ethics ref: KB42712021), we collected 309 recordings under various lighting conditions (5-11,800 lux). Additionally, we performed 300 measurements on 2 volunteers using three iPhone models (SE2, 13, 15Pro) at phone-face distances of 50-210 mm in dim lighting (275-350 lux). We developed adaptive algorithms to recognize the scene, adjust flash intensity, and provide real-time guidance for positioning. We compared measurements using fixed vs. adaptive settings.
The lighting study showed no significant correlation between pupil reactivity and illumination (Pearson's r=0.064, p=0.26). In the hardware/distance study, a linear mixed effects model showed that adaptive settings significantly reduced parameter dependence on distance: constriction amplitude (F(1,198)=9.27, p=0.0026), ratio (F(1,198)=11.02, p=0.0011), and PuRe score (F(1,198)=10.15, p=0.0017). Without adaptive settings, parameters varied significantly with distance (p<0.001), while adaptive settings eliminated these effects (p>0.5) with no significant device-related differences (F(2,99)=1.47, p=0.235).
The AI-powered pupillometer app with the invariant PuRe score shows consistent performance across varied lighting, distances, and phone models. This reliability enables effective pupillary assessment in diverse critical care and rehabilitation settings, promoting adoption for standardized neurological monitoring. The system's adaptability supports centralized health record data collection and consistent evaluation across care environments, regardless of measurement conditions or device models.
References:
Ong C, et al. (2019). Neurocrit Care, 30:316-321.
Bogucki A, et al. (2024). Front Neurol, 15:1363190. 10.3389/fneur.2024.1363190
Bogucki A, et al. (2024). i-Perception, 15(1):1-4.