Mammographic mass Classification using DL based ROI segmentation and ML based Classification

Mammography is said to be the initial examination for analysis of breast cancer masses in women at the age of 38 or above. Breast cancer masses may be benign or malignant. Various CAD systems have been designed to differentiate benign and malignant breast cancer masses. Breast cancer masses can be easily classified as benign or malignant considering their features. Features from mammographic images can be efficiently extracted using shape and texture features. Both the shape and texture features contribute a vital role in the classification of mammographic masses. Enormous Computer-assisted design systems have been employed for the classification of masses in mammographic images. In the present research work, the performance of the CAD system with DL based ROI segmentation and ML based mass classification of SFM mammographic images has been analyzed using 518 mammographic images from the DDSM dataset. DL based encode-decoder segmentation model ResNet50 is used for ROI segmentation. The shape and texture feature sets are extracted and concatenated, yielding a combined feature set to be fed to ML based classifiers namely ANFC-LH and PCA-SVM for binary classification of mammographic masses. The ANFC-LH and PCA-SVM classifier yields 73.0 % and 72.0 %, accuracy for the binary classification of SFM mass images.