Multilevel Deep Feature Generation Framework for Automated Detection of Retinal Abnormalities Using OCT Images

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) is crucial for diagnosing retinal disorders.
• Existing learning techniques for OCT image analysis have limitations.
• Automated and accurate detection of retinal disorders is essential.

Purpose of the Study:

• To develop a novel framework for automated retinal disorder detection using OCT images.
• To extract deep features from 18 pre-trained convolutional neural networks (CNNs).
• To achieve high classification performance for retinal disorders.

Main Methods:

• A three-phase framework: deep fused and multilevel feature extraction using 18 pre-trained CNNs and tent maximal pooling.
• Feature selection using iterative ReliefF (IRF).
• Classification using a quadratic support vector machine (QSVM).

Main Results:

• The framework achieved 97.40% classification accuracy on OCT dataset DB1.
• The framework achieved 100% classification accuracy on OCT dataset DB2.
• Demonstrated the effectiveness of deep feature extraction and selection for retinal disorder diagnosis.

Conclusions:

• The proposed framework successfully automates retinal disorder detection from OCT images.
• The combination of deep CNN features and optimized classification yields superior performance.
• This approach offers a promising tool for clinical ophthalmology and medical imaging analysis.