Detection of Retinal Abnormalities in OCT Images Using Wavelet Scattering Network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Optical Coherence Tomography (OCT) imaging is crucial for diagnosing retinal abnormalities.
• Current Computer Aided Diagnosis (CAD) methods often involve complex deep learning algorithms.
• Manual OCT B-scan analysis by ophthalmologists is time-consuming and prone to errors, especially in multiclass scenarios.

Purpose of the Study:

• To introduce a low-complexity CAD system for classifying retinal OCT images.
• To utilize a wavelet scattering network for identifying normal retinas and four specific pathologies: Central Serous Retinopathy (CSR), Macular Hole (MH), Age-related Macular Degeneration (AMD), and Diabetic Retinopathy (DR).

Main Methods:

• Employed a wavelet scattering network, a convolutional network utilizing predefined wavelets, for feature extraction.
• The network's filters are fixed, reducing processing time and complexity.
• Extracted features were classified using Principal Component Analysis (PCA).

Main Results:

• Achieved 97.4% accuracy in diagnosing abnormal retinas and 100% accuracy in identifying Diabetic Retinopathy (DR) from normal ones.
• Attained 84.2% accuracy in classifying OCT images into five classes (normal, CSR, MH, AMD, DR).
• Outperformed existing state-of-the-art methods, particularly those with high computational complexity.

Conclusions:

• The proposed wavelet scattering network offers an efficient and accurate CAD system for retinal OCT image classification.
• This method demonstrates potential for clinical application, reducing diagnostic time and improving accuracy, especially for complex cases.
• The system requires minimal data for learning, making it practical for clinical settings.