Spectral and Statistical Features in Fuzzy Neural Expert Machine for Colorectal Adenomas and Adenocarcinoma Classification

This paper presents a novel method which automatically detects differences in biopsy images of the colorectal polyps, extracts the required histopathology information through Fourier and statistical images analysis of the microscopic images and then classifies the cells into normal adenomas and malignant adenocarcinoma. The images are captured by a CCD camera from a laboratory microscope slide and store in computer using the .TIF format. The new system is implemented by fuzzifying image histopathological data. These are shape and texture descriptors calculated from the spectral analysis and greyscale statistical co-occurrence matrix analysis of the microscopic cell images, using the fuzzy neural network backpropagation classifier to differentiate the images. The novel system has been evaluated using 116 cancers and 88 normal colon polyp images collected from 44 normal patients and 58 cancer patients at random resulted in 96.5% classification accuracy. The breakthrough is that the algorithm is independent of the feature extraction procedure adopted; takes into consideration the gross and micro examination conducted by the pathologist and overcomes the sharpness of class characteristics associated with other classifiers algorithms.