Multimodal Breast Cancer Classification Using Wavelet Transform Features With GAN ‐Based Image Augmentation and Whale Optimization Algorithm for Feature Selection

Breast cancer is one of the most prevalent diseases among women. An automated system is necessary for the early diagnosis of the disease because manual diagnosis is time‐consuming and detection systems are scarce. Additionally, compared with a single imaging modality, using multiple modalities provides complementary information. In this work, a self‐collected multimodal, real‐time dataset comprising ultrasound and mammography images from the same patient has been utilized. Furthermore, a generative adversarial network (GAN) is used for image augmentation, and image features are extracted using wavelet‐transform‐based methods. Feature selection is carried out with the whale optimization algorithm (WOA), and multimodal classification is done with a long short‐term memory (LSTM) network. Using five‐fold cross‐validation, the proposed technique achieves an accuracy of 92.00%, a sensitivity of 98%, and an AUC of 0.978, which is higher than that of unimodal mammography or ultrasound systems. The results indicate that employing ultrasonography in conjunction with mammography will be beneficial to discriminate between benign and malignant tumors. The proposed method can be used efficiently by medical professionals, enabling early detection and timely treatment of breast cancer.