Multi-Scale Learning with Sparse Residual Network for Explainable Multi-Disease Diagnosis in OCT Images

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical coherence tomography (OCT) provides high-resolution, non-invasive imaging of biological tissues.
• Deep learning excels at diagnosing single retinal diseases from OCT images.
• Current methods struggle with multi-disease diagnosis in clinical OCT scans.

Purpose of the Study:

• To develop an effective deep learning approach for multi-disease diagnosis in OCT images.
• To enhance the discriminative capability and interpretability of disease prediction models.
• To reduce model complexity while maintaining high diagnostic performance.

Main Methods:

• A multi-scale learning (MSL) method was used to extract and fuse features from OCT images of varying sizes.
• A sparse residual network (SRN) was implemented, replacing large convolutional kernels with smaller ones to reduce complexity.
• The combined multi-scale sparse residual network was trained and evaluated on OCT datasets.

Main Results:

• The proposed multi-scale sparse residual network achieved 97.40% accuracy, 95.38% sensitivity, and 98.25% specificity.
• The method significantly outperformed existing approaches for multi-disease diagnosis in OCT images.
• Experimental results demonstrated the potential for improved, explainable diagnosis systems.

Conclusions:

• The developed multi-scale sparse residual network offers a robust solution for multi-disease diagnosis in OCT imaging.
• This approach enhances diagnostic accuracy and interpretability, aiding clinical decision-making.
• The findings pave the way for more advanced, explainable AI systems in ophthalmology.