Keywords: optical coherence tomography, pattern recognition, automatic tissue classification, computer-aided diagnostics, machine learning, image segmentation
Summary:Analysis of breast tissue images using a technique called LBP (Local Binary Pattern) is considered as a promising method for computer-aided detection of breast cancer cells. This technique calculates the statistical distribution of different patterns and uses them as feature vectors to represent textural differences. By removing the need for texture images to be highly repetitive, LBP enables improved tissue classification of OCT/OCM (Optical Coherence Tomography/Optical Coherence Microscopy) images. In this invention, the LBP method has been further refined into a process called ALBP (Average LBP) and BLBP (Block LBP). ALBP and BLBP features provide an enhanced encoding of the texture structure in a local neighborhood by looking at intensity differences among neighboring pixels and among certain blocks of pixels in the neighborhood. The ALBP feature compares the grayscale value of a neighbor pixel with the average grayscale value of all neighbors; in this way, ALBP can represent the intensity differences among neighbor pixels. The BLBP feature compares the average intensity values of pixels in blocks of certain shape around the center pixel, thus can represent more global intensity difference information that is not captured by the original LBP features. By integrating LBP features with the newly introduced two variants (ALBP and BLBP features), tissue classification accuracy can be significantly improved.