Wavelet Transform and Texton based Analysis for Detection of Benign and Malignant Masses.

Cancer has affected the human community to a large extent due to its low survival rate towards the end stage of the disease. It is asymptomatic in many cases during the initial stage. Thus the dependency on early diagnosis and regular check up increases manifold. Computer Aided Diagnostic Model is the need of the hour which will increase the diagnostic efficiency. A total of 400 images acquired from the Digital Database for Screening Mammography have been used here for analysis. This paper proposes a novel technique to differentiate benign and malignant breast lesions in mammograms using multiresolution analysis and Schmid Filter Bank, which were not reported earlier. A three level Haar wavelet decomposed image(L1, L2, L3) is obtained for each Region of Interest. In each level Texton based analysis is further investigated through Schmid filter bank. Statistical features and Haralick's Features are obtained from filter response and Gray Level Cooccurence Matrix respectively. Partition Membership Filter is further applied to the feature matrix for feature partitioning. The method shows maximum accuracy of 98.63% and Area under Curve of 0.981 using Random Forest Classifier and ten fold cross validation.