Automated macular pathology diagnosis in three-dimensional (3D) spectral domain optical coherence tomography (SD-OCT) images

A method of analyzing an optical coherence tomography (OCT) image of a retina, comprising:

• aligning a 2-dimensional (2D) slice of the OCT image to produce an approximately horizontal image of the retina;
• constructing an edge map based at least in part on the aligned 2D slice;
• determining at least one global representation based at least in part on the 2D slice or the edge map, wherein determining the at least one global representation comprises determining a (multi-scale) spatial division, wherein the (multi-scale) spatial division is a multi-scale spatial pyramid (MSSP);
• creating at least one local binary pattern (LBP)-based feature based at least in part on the at least one global representation;
• and
• classifying the 2D slice into one or more categories via one or more classifiers, wherein each category is associated with at least one ocular pathology, wherein the classifying is based at least in part on the at least one LBP-based feature.

A method of automated diagnosis of one or more macular pathologies, comprising:

• aligning an optical coherence tomography (OCT) image of a retina, comprising at least one of:
• detecting a majority of the retina structures by thresholding the image;
• removing noise from the image by applying a median filter;
• using morphological closing and opening to find an entire image of the retina;
• using a least-square curve fitting to fit a curve to the entire image of the retina;
• or
• warping the entire image of the retina to be approximately horizontal based at least in part on the curve;
• constructing one or more features based on the aligned image of the retina, comprising:
• creating an edge map based on the aligned image;
• constructing at least one (multi-scale) spatial division based on one or more of the aligned image or the edge map, wherein the at least one (multi-scale) spatial division comprises a multi-scale spatial pyramid (MSSP);
• and
• forming at least one local binary pattern (LBP)-based feature based at least in part on the at least one global representation;
• and
• classifying the image into a category for each of the one or more macular pathologies, wherein the classifying comprises employing a classifier to determine the category.