Enhancing Strabismus Diagnosis from Detection to Classification with Deep Learning

This project focuses on improving strabismus diagnosis by integrating detection and classification in a single deep learning pipeline. Strabismus is a condition in which the eyes fail to align properly, leading to visual imbalance. Traditional assessment depends on manual observation, which can be slow and inconsistent. The goal of this project is to create a structured, automated system that supports faster, clearer, and more reliable identification of eye misalignment.

The project uses a dataset containing labeled eye images and applies two modern object-detection models, YOLOv8 and YOLOv9, to build a complete diagnostic workflow. The system begins by detecting eye regions and then classifying them into normal or various forms of strabismus. A Flask-based backend manages model processing, while the interface is created using HTML, CSS, and JavaScript. Users can register, log in, upload images, and access prediction history through a simple and organized interface.

The design emphasizes efficiency, readable output, and structured analysis. By combining detection with classification, the system reduces dependency on manual inspection and provides consistent results. Features such as user authentication, prediction tracking, and integrated model usage create a complete diagnostic assistant.

This project demonstrates how optimized deep learning methods and simple web technologies can work together to support better screening and analysis of strabismus, offering an accessible and repeatable approach to automated eye-alignment evaluation.

Keywords: strabismus, deep learning, YOLOv8, YOLOv9, detection, classification, eye alignment, computer vision, Flask, automation.