Five-dimensional analysis of multi-contrast Jones matrix tomography of posterior eye

Abstract

Pixel clustering algorithm tailored to multi-contrast Jones matrix based optical coherence tomography (MC-JMT) is demonstrated. This algorithm clusters multiple pixels of MC-JMT in a five-dimensional (5-D) feature space which comprises dimensions of lateral space, axial space, logarithmic scattering OCT intensity, squared power of Doppler shift and degree of polarization uniformity. This 5-D clustering provides clusters of pixels, so called as superpixels. The superpixels are utilized as local regions for pixels averaging. The averaging decreases the noise in the measurement as preserving structural details of the sample. A simple decision-tree algorithm is applied to classified superpixels into some tissue types. This classification process successfully segments tissues of a human posterior eye.