Automatic diagnosis of abnormal macula in retinal optical coherence tomography images using wavelet-based...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate classification of macular abnormalities from 3D OCT scans is crucial for timely diagnosis and treatment.
• Current methods often require extensive preprocessing steps like denoising and segmentation.
• Automated analysis can improve efficiency and consistency in interpreting OCT data.

Purpose of the Study:

• To develop a fully automatic algorithm for classifying 3D OCT scans.
• To differentiate between normal macula and eyes with abnormalities such as diabetic macular edema (DME) and age-related macular degeneration (AMD).
• To eliminate the need for manual preprocessing steps.

Main Methods:

• A two-stage classification scheme employing adaptive feature learning and diagnostic scoring.
• Utilized a wavelet-based convolutional neural network (CNN) for feature extraction in the spatial-frequency domain.
• Evaluated on two distinct SD-OCT datasets using fivefold cross-validation.

Main Results:

• Achieved an average precision of 99.33% for two-class classification (normal vs. DME) on the first dataset.
• Obtained an average precision of 98.67% for three-class classification (AMD, DME, normal) on the second dataset.
• Demonstrated high performance without requiring denoising, segmentation, or alignment.

Conclusions:

• The proposed automatic algorithm effectively classifies 3D OCT scans for macular abnormalities.
• The method offers a robust and efficient alternative to traditional, multi-step analysis pipelines.
• High precision in classifying DME and AMD highlights its clinical potential.