Mammography image classification by convolution features and ensemble Learning Using Random Forest with Boosting approach

Abstract

Medical images comprising of automatic tissue classification represents a significant stage both in the detection and diagnosis of pathology. Here, we usually deal with the mammographic images and present a novel framework based on deep learning mechanism for region-based classification into semantically coherent type of tissues. On comparing with support vector machine (RANDOM FOREST(RF)) classifiers dependent on the features extracted utilizing Convolutional Neural Network (CNN) and our earlier computer extricated tumour features includes the task for recognizing malignant and benign lesions. A mechanism based on Five-fold cross validation was led with the region under receiver operating characteristic (ROC) curve as the performance-based metric. The proposed technique CNN based RANDOM FOREST(RF) to study discriminative features on natural basis. The proposed methodology is engaged with a concise database via training the CNN-RANDOM FOREST(RF) in an overlapped patch-wise way so as to quicken the pixel-wise class-based prediction. Here we utilize convolutional layers rather than the traditional fully connected layers. This methodology altogether brings about fast computation, while protecting the accuracy of classification. The proposed strategy was tried on marked mammographic images and exhibits promising results in terms of image segmentation along with classification of tissues.