Abstract
This article presents the design, implementation, and validation of a low-cost embedded system for preliminary strabismus screening, based on computer vision and deep learning. The hardware integrates a Raspberry Pi 4, a USB camera, and a 3D-printed chin rest to ensure consistent facial positioning. The software, developed in Python using PyQt5 and OpenCV, incorporates a NASNetLarge convolutional neural network converted to TensorFlow Lite for real-time inference. The graphical interface allows users to capture or upload images, perform automated analysis, generate diagnostic PDF reports, and access a gamified treatment module. Functional validation included a proprietary dataset of 27 images, achieving a 96.30 % classification accuracy. Additionally, a stratified 10-fold cross-validation on a balanced dataset of 1000 images yielded an average accuracy of 95.6 % with strong generalization metrics (F1-score, precision, and recall above 94 %). A novel treatment validation mechanism was implemented by analyzing pupil-to-stimulus distance frame-by-frame, confirming reliable eye tracking and the system's potential for detecting microstrabismus. This open-source, portable prototype is suitable for community health screening and educational use, particularly in low-resource settings.