Automatic and robust segmentation of endoscopic OCT images and optical staining

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Optical Coherence Tomography

Background:

• Accurate segmentation of tissue layers in endoscopic optical coherence tomography (OCT) images is crucial for quantitative analysis and clinical diagnosis.
• Current methods may lack robustness or require manual intervention, limiting real-time applications.

Purpose of the Study:

• To develop a generic, automatic method for segmenting endoscopic OCT images.
• To enhance signal-to-noise ratio (SNR) and contrast for improved layer differentiation.
• To enable in vivo optical staining histology and quantitative analysis of tissue geometry.

Main Methods:

• Image de-noising and smoothing using L-L norm minimization.
• Formulation of cost graphs based on vertical image gradients.
• Tissue-layer segmentation utilizing a shortest path search algorithm.

Main Results:

• Robust and automatic identification of all five esophageal layers in guinea pig endoscopic OCT images.
• Demonstrated high segmentation accuracy, enabling in vivo optical staining histology.
• Facilitated quantitative analysis of tissue geometric properties.

Conclusions:

• The developed method provides a generic and accurate approach for endoscopic OCT image segmentation.
• This technique supports real-time, in vivo histological analysis and quantitative tissue assessment.
• Potential applications include studying pathologies and aiding in clinical diagnosis.