Automated strabismus detection and classification using deep learning analysis of facial images.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Strabismus, or eye misalignment, is a prevalent condition requiring early detection for effective management.
• Accurate classification of strabismus is crucial to prevent long-term visual complications.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated strabismus identification and classification using facial images.
• To assess the model's performance in both binary (strabismus vs. normal) and multi-class (deviation angle) classification tasks.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) for image analysis.
• Trained and validated the model on datasets comprising 4,257 images for binary classification and 622 images for multi-class classification.
• Employed five-fold cross-validation and evaluated performance using accuracy, sensitivity, F1-score, and recall.

Main Results:

• The deep learning model achieved 86.38% accuracy for binary strabismus classification.
• The model demonstrated 92.7% accuracy for multi-class classification of strabismus deviation angles.
• Performance metrics confirmed the model's effectiveness in identifying and classifying strabismus.

Conclusions:

• The proposed deep learning approach shows significant potential for assisting healthcare professionals in the early detection of strabismus.
• This technology can aid in strabismus treatment planning, potentially improving patient outcomes.
• Automated analysis of facial images offers a promising tool for ophthalmological diagnostics.