Classification of Choroidal Neovascularization and Diabetic Macular Edema Based on Feature Extraction from Optical...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) is crucial for diagnosing macular diseases.
• Diabetic macular edema (DME) and choroidal neovascularization (CNV) are leading causes of vision loss.
• Biomarker identification from OCT images is key for clinical assessment.

Purpose of the Study:

• To develop and validate a Python algorithm for extracting OCT biomarkers.
• To assess the algorithm's performance in classifying normal, DME, and CNV OCT images.
• To explore the potential for integrating machine learning into automated macular disease diagnosis.

Main Methods:

• A Python algorithm was created to extract biomarker-associated features from OCT images.
• The algorithm was applied to a pre-labeled dataset of normal, DME, and CNV images.
• LightGBM machine learning model was used for image classification based on extracted features.

Main Results:

• Extracted features showed distributions consistent with existing literature.
• The LightGBM classifier achieved 91% accuracy in diagnosing macular diseases.
• An area under the receiver operating characteristic curve of 98% was obtained.

Conclusions:

• The developed algorithm shows promise for advanced feature extraction in OCT imaging.
• This approach can contribute to the development of machine learning-based diagnostic tools for macular diseases.
• Potential for integration into automated patient screening systems for early detection and management.