Mammographic Cancer Identification By Using DCNN (Deep Convolutional Neural Network)- A review

It is recently observed that, most of the women are suffered from the bosom cancer disease all over the world. bosom cancer has become one of the major health challenges. Also it is an extremely difficult and time consuming task for the pathologists to analyse the complicated structure of bosom cancer tumor. During diagnosis, the examination of histopathological images plays an important role. For the feature extraction of these histopathological images, the conventional feature removal method extracts only small level features. Therefore it is necessary to design a proper feature extraction method to extract important features from these histopathological pictures. To extract the high level features automatically from an image, deep learning methods are used. Therefore, in this dissertation work, we used a DCNN (Deep Convolutional Neural Network) to extract the different features of histopathological images. In the first step, we have developed two architectures namely Inception-V3 and Inception-ResNet V2. These two architectures are designed to solve the multi-class and binary issues of classification of histopathological images of bosom cancer. The imbalance problem of histopathological images is present in a different subclasses, where it can be removed by rotating the image in clockwise - anticlockwise, up - down, left - right and in different degrees and for this purpose Ductal Carcinoma is considered as a baseline. The experimental results of this dissertation work shows that the categorization of histopathological pictures of bosom cancer are more superior to the existing methods. However, the Inception-ResNet V2 network executes unverified examination of the histopathological pictures for bosom cancer correctly, and it also correctly extracts the different features from the images. These extracted features are converted into low dimensional space by using auto-encoder network for the analysis of clustering of the image.