The Evolution of Breast Cancer Detection: A Review of Imaging, Machine Learning, and Multimodal Strategies

ABSTRACT

Breast cancer is still a serious problem in the world arena, where its early and prompt detection is the most important factor in improving patient prognosis and survival. The use of traditional diagnostic techniques, such as imaging (e.g., mammography and ultrasound) and subsequent histopathological examination, is the mainstay, which, however, is hindered by a variety of limitations and is less sensitive in certain instances (e.g., dense breast tissue). Computer‑aided diagnosis (CAD) systems based on machine learning (ML) and deep learning (DL) approaches have become revolutionary technologies to tackle these issues. The review is a comparative analysis of all the various methodologies involved in detecting breast cancer. It first analyzes the conventional imaging methods and their shortcomings. It then reviews the use of classical ML algorithms (e.g., support vector machines, random forests) mostly on structured clinical data. The main idea of the review is the popularity of DL to evaluate a number of medical images, such as histopathology slides, ultrasound, and mammograms, in particular, convolutional neural networks (CNNs). Finally, the review examines the next generation of hybrid approaches, such as ensemble learning, data fusion architectures, and combinations of multimodal data, such as genomics, pathomics, and metabolomics. This paper assesses the performance, strengths, and limitations of these coexisting strategies by synthesizing recent findings, noting that there are still some persistent challenges in terms of data scarcity, model interpretability, and clinical translation. The review has ended with a summary of the present situation and the future of breast cancer diagnostics.