Development and validation of a multi-stage self-supervised learning model for optical coherence tomography image...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate classification of optical coherence tomography (OCT) images is crucial for diagnosing eye diseases.
• Traditional supervised learning models require large, expertly annotated datasets, which are costly and time-consuming to acquire.
• Self-supervised learning offers a promising alternative to reduce reliance on labeled data.

Purpose of the Study:

• To develop and validate a novel multi-stage self-supervised learning model for OCT image classification.
• To assess the model's performance against conventional supervised and self-supervised methods.
• To evaluate the model's robustness and diagnostic accuracy, particularly under conditions of limited labeled data.

Main Methods:

• A multi-stage self-supervised learning framework was developed.
• The model was trained and validated on a private dataset and three public OCT image datasets.
• Performance was evaluated using internal, external, and clinical validation, including Grad-CAM for interpretability and subsampling analyses for robustness.

Main Results:

• The proposed model achieved state-of-the-art results on public datasets, outperforming conventional methods.
• In clinical validation with limited data, the model demonstrated up to 17.50% higher accuracy and 17.53% higher macro F-1 score than a supervised model.
• Grad-CAM analysis provided insights into the model's decision-making process.

Conclusions:

• The multi-stage self-supervised learning model effectively addresses the challenge of limited labeled data in OCT image classification.
• The model shows significant potential for improving diagnostic accuracy and efficiency in ophthalmology.
• Availability of source code and pre-trained models facilitates clinical adoption and workflow integration.