A texture based approach for automatic identification of benign and malignant tumor from FNAC images

Cancer is one of the most destructive diseases which if not detected in time, will surely lead to death. About 12 million people will be died due to cancer by 2030 as per the statistics, provided by the World Health Organization (WHO). Thus a big challenge and area of research emerges in front of both the medical practitioner and scientific researcher to fight against cancers. When a patient is suspected for the presence of malignant tumor they are advised for FNAC (Fine Needle Aspiration Cytology) test where specimens of cells can be taken in minimally invasive way with, e.g., tiny needles, with or without syringes. One of the main drawbacks of cytopathological diagnosis is the time required for an expert to visually inspect a specimen under a microscope, in search of malignant or suspicious cells and manually select them for further analysis. The present work tried to device an automated computer-aided diagnostic system specifically to reduce time and provide `second opinion´ for pathologists in making diagnosis. A database of 100 FNAC images were taken on which k-fold cross-validation was performed, where k varied, for the diagnosis of malignancy. Initially, elimination of cytoplasm from the images consisting of multiple cells was done by performing saturation threshold segmentation and from the segmented nucleus boundary, meaningful texture and shape describing features are calculated using GLCM and LBP algorithms. The outcome of segmentation followed by feature extraction was tested by using the Logistic classifier which is a machine learning algorithm. The achieved diagnostic accuracy is 86%, when features obtained by combining GLCM and LBP methods, are used for classification.