CNN-FS-IFuzzy: A new enhanced learning model enabled by adaptive tumor segmentation for breast cancer diagnosis using 3D mammogram images

• Published in: Knowledge-based systems Vol. 288; p. 111443
• Main Authors: Umamaheswari, Thippaluru, Murali Mohanbabu, Y.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 15.03.2024
• Subjects: 3D mammogram adaptive thresholding with region growing fusion model; Breast cancer diagnosis; Cat-rider swarm optimization algorithm; Convolution Neural Network; Improved Fuzzy-based learning model; optimal Feature Selection; optimal Feature Selection; Convolution Neural Network; Cat-rider swarm optimization algorithm; 3D mammogram adaptive thresholding with region growing fusion model; Breast cancer diagnosis; Improved Fuzzy-based learning model
• ISSN: 0950-7051; 1872-7409
• DOI: 10.1016/j.knosys.2024.111443

In recent years, breast cancer has caused death among women around the world. Detecting breast cancer in the early stage helps to eradicate the survival rate to aid accurate medical treatments. The early detection of breast cancer is necessary, and thus, automatic identification of abnormalities in the breast is more essential, which increases the chances of survival. For this purpose, mammography is chosen in this research work for detecting and screening breast cancer at an earlier stage. Moreover, early detection of breast cancers can be done with the help of examining digital mammography, in which 3-D breast mammography was chosen as a complementary modality. The Computer-Aided Detection (CAD) system is proposed to facilitate the analysis of 3-D breast mammography images, in which segmentation of breast cancer plays a major role in extracting the features and temporal evaluations. Though, the automatic segmentation of masses is a complicated task due to the large diversity in texture, size, and shape of these 3-D objects. The major scope of the research is to enhance the breast cancer detection model using 3D mammogram images enabled by deep learning. As an initial step, the pre-processing of the images is handled by the median filtering and image scaling model. Consequently, the image segmentation is performed by the Adaptive Thresholding with Region Growing Fusion Model (AT-RGFM). Here, the adaptive concept of tumor segmentation is done by the hybrid Cat Swarm Optimization (CSO) and Rider Optimization Algorithm (ROA), termed the Cat-Rider Swarm Optimization Algorithm (C-RSOA). Although, the segmented tumor images are subjected to the deep classification called CNN-FS-IFuzzy learning model, in which the deep features are extracted to the pooling layer of the Convolution Neural Network (CNN), followed by optimal Feature Selection (FS) by hybrid C-RSOA, and extends the classification by Improved Fuzzy (IF)-based learning model. Here, the C-RSOA will perform the membership limit optimization of fuzzy, which intends to attain high detection accuracy. The validation results prove that the proposed segmentation and classification perform better than the existing models.