Enhancing Pupil Detection Accuracy for Medical Diagnostics using PINNs Combined with VGG19

Medical diagnostics and human-computer interaction need accurate pupil detection yet existing methods fail under various lighting conditions and occluded environments. The proposed framework implements an integration of Physics-Informed Neural Networks (PINNs) using VGG19 features extraction capabilities to overcome system limitations. Integrating physical constraints like pupil circularity and illumination effects into neural network learning mechanisms, PINNs ensure stronger robustness and better generalization abilities. Testing on LPW and OpenEDS datasets yielded performance measures of IoU 0.92 with F1-score 96.4% and MAE 1.23 pixels. The real-time detection capabilities of this solution were confirmed through average detection times below 15 milliseconds. Testing from advanced methods demonstrated the proposed system earned top rankings in all performance criteria. The proposed framework resolves pupil detection difficulties, creating possibilities for improved medical diagnostics and eye-tracking technologies during real-world application development.