Colon Cancer Prediction On Different Magnified Colon Biopsy Images

Colon Cancer detection is an important task for the histopathologist as they have to analyze morphology of the images at different magnifications thereby leading to intra and inter observer variability. Thus an automation is needed to detect the colon cancer for all the magnifications. This paper focus on the colon cancer detection on the colon biopsy images at different magnifications. First the images are converted to its HSV color space and further processing are performed on the saturation component. Features are the coefficients obtained from Dual-tree and double-density 2-D wavelet transform. In the last phase classification is performed with the Random Forest classifier. The proposed system was evaluated on colon biopsy images collected from Aster Medcity at different magnifications 10X, 20X, 40X and combined magnifications where an accuracy of 85.4% is obtained for the combined magnifications.