Innovative AI-based system for precision diagnosis of childhood strabismus incorporating gaze tracking and real-time correction feedback.

PurposeStrabismus is a common pediatric eye disorder that can lead to developmental and psychosocial consequences if not treated promptly. Traditional diagnostic methods often depend on clinician expertise, which can result in variability and delayed intervention.MethodsA prospective diagnostic accuracy trial was conducted with 250 children aged 3 to 10 years. Eye-tracking and image data were collected using a Tobii Pro Fusion device and analyzed using a dual-stream deep learning model (ResNet-50 combined with a 1D-CNN). The diagnostic performance was compared against pediatric ophthalmologists and existing AI methods. A real-time feedback module was implemented to provide vergence and anti-suppression training, enabling assessment of its therapeutic impact.ResultsThe AI system achieved a binary accuracy of 96.8% (AUC = 0.992) and a multi-class accuracy of 92.5% (AUC = 0.983), demonstrating agreement with gold-standard diagnoses (κ = 0.931). Ablation studies indicated that the multimodal fusion performed better than unimodal baselines (p < 0.01). Therapeutic evaluation showed a 30.1% reduction in deviation angle, a 17% improvement in fusion, and high adherence rates (87%). Diagnostic efficiency improved, with evaluation times reduced by 62.1% (p < 0.001), and parental satisfaction scores increased (p < 0.01).ConclusionThe findings indicate that the proposed multimodal AI system provides diagnostic accuracy comparable to specialist evaluation, along with measurable therapeutic benefits and improved operational efficiency. By integrating diagnostic and rehabilitative functions, the system presents a patient-centered approach to pediatric strabismus care with potential for application in clinical and resource-limited settings.