Modified U-Net and Convolutional Networks for Breast Cancer Segmentation and Classification with New Texture Descriptors

Abstract

Effective techniques for detection and classification are necessary to enhance the results of patients with common fatal diseases like breast cancer. With benefits including increased effectiveness and precision, DL and ML methods are currently being used as effective tools in the breast cancer diagnosis process. However, the efficiency and stability constraints of current approaches highlight the necessity for additional study. Therefore, this work suggested a new model of modified U-Net and convolutional networks for Breast Cancer (BC) segmentation and classification with new texture descriptors and the proposed model is named “U-TexBCNet”. First, mammography images are preprocessed by enhancing the image using the CLAHE filter and after inputting the preprocessed image into the modified U-Net-based segmentation for the segmentation process, the modified U-Net segmentation is named as “HexpNet”. Once the segmentation process is completed, the feature extraction process takes place. Here, the features include shape feature, color feature, VGG-16 feature, Inception V3 feature, GoogLeNet and modified LGTP feature extracted from the segmented image. Then, the extracted features are inputted into the hybrid classification model including modified CNN and LSTM. Lastly, the outcome from the classifiers (Modified CNN and LSTM) is averaged together to get the outcome of the proposed model. Ultimately, the results of the simulation are evaluated using both positive and negative criteria by contrasting them with the conventional models.