Cost-Saving Data-Driven Diabetic Retinopathy Prediction via a Sampling-Empowered Incremental Learning Approach.

Area of Science:

• Medical Informatics
• Machine Learning
• Ophthalmology

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss, necessitating early detection.
• Machine learning models using electronic health records (EHR) show promise for DR prediction.
• Periodic EHR updates require efficient model retraining strategies.

Purpose of the Study:

• To develop an incremental learning framework for DR prediction models.
• To enable continuous learning from new EHR data while retaining prior knowledge.
• To reduce the cost and complexity of model retraining.

Main Methods:

• Proposed an incremental learning framework integrating a weighted sampling strategy.
• Tested the approach on various classification models for DR prediction.
• Evaluated model performance on periodically updated EHR data.

Main Results:

• The sampling-empowered incremental learning approach demonstrated higher efficiency.
• The framework achieved more accurate DR prediction compared to traditional methods.
• Effectively mitigated challenges of periodically updated EHR databases.

Conclusions:

• The proposed incremental learning framework offers a cost-effective solution for DR prediction.
• Enables healthcare providers to maintain high prediction accuracy with updated EHR data.
• Facilitates timely diagnosis and treatment of diabetic retinopathy.