Classification-based framework for binarization on mice eye image in vivo with optical coherence tomography.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Ophthalmology

Background:

• Optical coherence tomography (OCT) angiography is a key medical imaging technique.
• Binarization is crucial for quantitative analysis in OCT, particularly for ocular imaging.
• Existing methods struggle with limited training samples and low-contrast OCT images.

Purpose of the Study:

• To develop a novel binarization framework for low-intensity OCT images.
• To address challenges of few training samples and contrast limitations in OCT binarization.
• To introduce an open, classification-based framework named specific-patch SVM (SPSVM).

Main Methods:

• A two-phase approach: model training and binarization threshold determination.
• Extraction of target and background patches from limited training samples.
• Application of Principal Component Analysis (PCA) for dimensionality reduction and feature extraction.
• Training a Support Vector Machine (SVM) classification model using extracted patch features.
• Utilizing the learned subspace and SVM model for patch-wise binarization thresholding in testing.

Main Results:

• The proposed SPSVM framework demonstrates effectiveness in binarizing low-intensity OCT images.
• Successful handling of datasets with few training samples and low contrast.
• Acquisition and public release of a new OCT mice eye (OCT-ME) database.
• Extensive experiments validated the performance of the SPSVM framework.

Conclusions:

• The SPSVM framework offers a robust solution for binarizing challenging OCT images.
• The method improves quantitative analysis in OCT by overcoming data limitations.
• The publicly available OCT-ME database facilitates further research in OCT image analysis.