Deep learning-based classification of multiple fundus diseases using ultra-widefield images

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate diagnosis of fundus diseases is crucial for effective treatment.
• Ultra-widefield (UWF) imaging provides a comprehensive view of the retina.
• Current diagnostic methods can be time-consuming and require expert interpretation.

Purpose of the Study:

• To develop a hybrid deep learning model for classifying multiple fundus diseases using UWF images.
• To improve the diagnostic efficiency and accuracy of ophthalmic disease detection.
• To provide an auxiliary tool for clinical decision-making in ophthalmology.

Main Methods:

• A retrospective study involving 10,612 UWF fundus images covering 16 fundus diseases.
• Utilized a hybrid model combining DenseNet121 feature extraction with an XGBoost classifier.
• Employed Gradient-weighted Class Activation Mapping (Grad-CAM) for model visualization and performance evaluation using accuracy, precision, recall, F1 score, and AUC-ROC.

Main Results:

• The model achieved high diagnostic performance, with AUC values exceeding 0.975 for common diseases and 0.970 for rare diseases.
• Accuracy rates were above 0.980 for common diseases and surpassed 0.998 for rare diseases.
• Grad-CAM visualizations confirmed alignment with clinical pathological features, and the model outperformed ophthalmologists in diagnostic accuracy.

Conclusions:

• The developed deep learning model effectively classifies multiple ophthalmic diseases from UWF images.
• This AI tool shows potential for enhancing clinical diagnostic efficiency and optimizing ophthalmologist workflows.
• The model promises to improve the quality of patient care in ophthalmology.