Breast cancer Classification with Multi-Fusion Technique and Correlation Analysis

Abstract

Breast cancer is responsible for the deaths of the vast majority of women who succumb to the disease. By detecting and treating the disease at an earlier stage, it is feasible to reduce the mortality rate associated with breast cancer. Mammography is the technique of breast cancer screening that has received the most amount of approval from the medical community. Imaging of the ipsilateral projections of the bilateral (right and left) breasts, also known as mediolateral oblique (MLO) and craniocaudal (CC) views, is often necessary for this surgery. This imaging technique is also known as the craniocaudal projection. Sonography, which is also known as ultrasound imaging, is used in combination with mammography during the diagnostic phase (which occurs after the screening phase) to offer a more accurate examination of any abnormalities that may have been detected. Radiologists may be able to make a more precise diagnosis of breast cancer by carrying out an objective assessment with the assistance of CAD systems. CAD is an abbreviation that stands for computer-aided detection and diagnosis. Researchers have proposed computer-aided design (CAD) systems as a viable technique for increasing system performance. These CAD systems take information from a variety of sources and combine it into a single database. In the majority of occurrences, this necessitates the inclusion of qualities or evaluations that were collected from a wide range of information sources. Fusion of choices is effective when dealing with sources that are statistically independent, while fusion of characteristics is good when dealing with sources that have a significant degree of correlation with one another. However, sources often contain a mix of information that is associated with one another as well as information that is independent of one another; as a consequence, none of these approaches is the greatest choice available to choose from. The development of optimal fusion strategies for Multiview and multimodal breast CAD systems is the major focus of this thesis. Canonical correlation analysis is the name of the statistical approach that serves as the foundation for these tactics (CCA). The CCA algorithm alters two multivariate datasets in such a manner as to maximize the correlation that already exists between them. This, in turn, optimizes the feature fusion that occurs after the CCA method has been applied. On the other hand, the performance of benchmark fusion schemes that combine all three sources of information is only at most equivalent to the performance of benchmark schemes that fuse two information sources. In addition, the performance of benchmark fusion schemes that combine all three sources of information is worse than the performance of CCA-based feature fusion schemes that combine two sources of information. This indicates that even if increasing the number of sources could bring new information, only a fusion approach that is optimized to exploit its maximum potential would be able to make the most of this extra data. In conclusion, the CCA-based fusion schemes exhibit robustness when tested against a wide array of performance indicators, datasets, information sources, and diagnostic tasks that are related to the diagnosis of breast cancer. The benchmark fusion techniques, on the other hand, do not demonstrate this resilience.