Cascaded Deep CNN Network for Breast Cancer Screening from Ultrasound Images Using Differential Evolution Algorithm

Because of poor blood flow and susceptibility to infections, breast cancer can be challenging to heal and, if not treated, might cause major problems, including limb amputation and a lowered quality of life. Though several systems exist to identify breast cancer, few combine machine learning (ML), deep learning (DL), and optimization strategies. This work presents a novel method leveraging complex algorithms to precisely identify breast cancer from ultrasound images. The study utilizes a dataset from a secondary source, categorizing it into three identical classes: normal (Class 0), benign (Class 1), and malignant (Class 2). It extracts features using pre-trained convolutional neural networks (CNNs), optimizes these features using the Differential Evolution Algorithm (DEA), and classifies images using standard machine learning algorithms. To show how well it detects breast cancer, the approach combines seven conventional ML classifiers with ten pre-trained CNN models. Using DenseNet121 + DEA + Xtreme Gradient Boosting (XGB) Classifier, DEA chooses important features derived by CNNs with a high accuracy of 97.76%. However, the proposed method can be very effective for real-time breast cancer screening using ultrasound images.