DEVELOPMENT AND VALIDATION OF AN EXPLAINABLE ARTIFICIAL INTELLIGENCE FRAMEWORK FOR MACULAR DISEASE DIAGNOSIS BASED ON...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Macular diseases are a leading cause of vision loss.
• Accurate diagnosis relies on expert interpretation of optical coherence tomography (OCT) images.
• Identifying retinal lesions is crucial for diagnosis.

Purpose of the Study:

• To develop and validate an AI framework for retinal lesion identification and macular disease diagnosis using OCT images.
• To achieve explainable disease diagnosis at the eye level.
• To compare the AI framework's performance against human experts.

Main Methods:

• A two-stage framework was developed using deep learning for lesion classification and random forests for disease diagnosis.
• The framework was trained and tested on 26,815 OCT images from 865 eyes, with expert-labeled retinal lesions and macular diseases.
• Performance was evaluated against human expert diagnoses.

Main Results:

• The deep learning model achieved an AUC of 0.978 for lesion classification.
• The random forest model accurately diagnosed macular diseases with a 0% error rate, matching one expert and outperforming three others.
• Lesion detection in the macular region significantly contributed to diagnosis.

Conclusions:

• The integrated AI framework demonstrates high accuracy and interpretability in retinal lesion classification and macular disease diagnosis from OCT images.
• This AI tool has the potential to assist clinicians in diagnosing macular diseases.
• Explainable AI in ophthalmology can enhance diagnostic confidence and efficiency.