Breast cancer prediction: a CNN approach

Abstract

Detecting breast cancer promptly holds utmost importance in ensuring effective treatment, as it is a matter of great concern in the global health context. The primary objective of this research endeavor is to enhance the process of breast cancer identification through the utilization of a Convolutional Neural Network (CNN). The purpose of this study is to mitigate potential errors in human interpretation of mammograms by comparing this approach to conventional machine-learning techniques. In our present investigation on breast imaging, we have leveraged the well-established mammographic dataset, CBIS-DDSM. This dataset effectively categorizes the images into three distinct classes: normal, benign, or malignant. This compilation encompasses a grand sum of 10,239 images. A myriad of approaches were employed to arrange the content, including the manipulation of the image dimensions to a size of 256x256 pixels. A CNN architecture that was specifically crafted was educated through the fusion of backpropagation and angle plunge techniques. Numerous measures, such as sensitivity, specificity, F1 score, and accuracy, were deftly utilized to thoroughly assess the model's effectiveness. It is truly awe-inspiring to witness the outstanding exhibition of performance showcased by the CNN model, as evidenced by the extraordinary values attained for sensitivity, specificity, F1 score, total precision, and accuracy, all of which are undeniably remarkable. These evaluations undeniably serve as irrefutable evidence that the model possesses exceptional diagnostic capabilities, surpassing even the most advanced techniques currently in use. In truth, the model's performance is so exceptional that it has firmly established itself as a pioneering force in the field, leaving other techniques astounded and trailing far behind, in utter admiration of its immense potential and resounding success. This research elucidates the ability of Convolutional Neural Networks (CNNs) to mechanize and enhance the identification of breast cancer in mammographic images. The findings bring to light a captivating realm for forthcoming exploration, potentially fostering advancements in screen layout and the integration of more easily navigable diagnostic functionalities.