Neurosymbolic AI Framework for Explainable Retinal Disease Classification From OCT Images.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate classification of retinal diseases is crucial for effective patient treatment.
• Traditional deep learning models face challenges with imbalanced datasets and lack interpretability in ophthalmology.

Purpose of the Study:

• To develop and evaluate a neurosymbolic framework for improved classification of retinal diseases.
• To enhance diagnostic performance and clinical interpretability in ophthalmology using AI.

Main Methods:

• A hybrid neurosymbolic framework integrating a convolutional neural network (CNN) with a symbolic reasoning layer was proposed.
• 10,846 optical coherence tomography images across seven diagnostic classes were analyzed.

Main Results:

• The neurosymbolic model achieved superior performance on both internal (macro-F1: 0.81) and external (macro-F1: 0.78) datasets compared to a standalone CNN.
• The model demonstrated improved macro-precision and recall, outperforming the CNN in diagnostic accuracy.

Conclusions:

• The proposed neurosymbolic framework offers a unified paradigm coupling symbolic reasoning with CNNs for enhanced diagnostic performance.
• This approach provides transparent, clinically interpretable decisions, particularly beneficial for rare and complex retinal conditions.