Near real-time classification of optical coherence tomography data using principal components fed linear discriminant...

Area of Science:

• Biomedical Optics
• Image Analysis
• Machine Learning

Background:

• Optical Coherence Tomography (OCT) is a valuable imaging modality.
• Automated data analysis is crucial for efficient OCT interpretation.
• Developing robust classification algorithms for OCT data is an ongoing challenge.

Purpose of the Study:

• To design and validate an automated prediction algorithm for Optical Coherence Tomography (OCT) data.
• To assess the algorithm's classification accuracy on diverse sample types.
• To demonstrate the feasibility of near real-time OCT data processing and classification.

Main Methods:

• Development of an automated preprocessing and classification algorithm for OCT images.
• Utilized surface recognition and normalization techniques for A-scan data reduction.
• Employed principal component analysis (PCA) and linear discriminant analysis (LDA) for classification.
• Validated the algorithm on a dataset of vegetables and porcine tissues comprising nine groups.

Main Results:

• The algorithm achieved 82% correct classification after cross-validation on a nine-group dataset.
• Higher accuracy rates were observed for datasets with fewer groups.
• The algorithm successfully distinguished between visually similar groups.
• Near real-time processing at 60,000 A-scans/min was achieved on standard hardware.

Conclusions:

• Automated OCT data classification is achievable with high accuracy using PCA and LDA.
• Surface normalization is a critical step for effective A-scan data reduction and classification.
• The algorithm shows promise for applications requiring rapid analysis of OCT data, including potential clinical settings, though further validation is needed.