FUSION VIEW-NET: DUAL-VIEW DEEP LEARNING FOR ROBUST MAMMOGRAPHIC BREAST CANCER CLASSIFICATION
Beibit Abdikenov, Tomiris Zhaksylyk, Aruzhan Imasheva, Yerzhan Orazayev, Danara Suleimenova · Scientific Journal of Astana IT University · 2025
Breast cancer is still one of the top causes of cancer-related death for women globally, and better patient outcomes depend on early identification. Although mammography is the main imaging modality used for screening, the delicate nature of early clinical symptoms and inter-reader variability sometimes compromise diagnostic accuracy. We examine the application of deep convolutional neural networks (CNNs) to automated classification of mammogram images in this work. FusionView-Net (FV-Net) is also presented, a novel dual-view integration framework that combines data from mediolateral oblique (MLO) and craniocaudal (CC) views to improve diagnostic precision. To produce a more comprehensive depiction of the breast tissue than conventional single-view methods, FV-Net combines contextual and spatial data from both standard perspectives. Two publicly available mammography datasets, which have been properly divided to allow for both seen-unseen data configurations and cross-dataset generalization testing, are used to assess the approach. A variety of CNN architectures are evaluated on separate and combined datasets, including ResNet18 and a specially created CNN. Findings indicate that FV-Net significantly increases model robustness and classification accuracy, as evidenced by consistently better F1 scores and ROC AUC values, especially when combined with ResNet18 and the custom CNN. The necessity for flexible models in actual clinical settings is shown by generalization studies, which further highlight the significance of dataset diversity by showing a noticeable drop in performance when domain shifts are present. Our results demonstrate how well multi-view fusion works for CNN-based mammography classification and provide useful guidance for choosing architectures and training methods. The development of trustworthy, broadly applicable AI technologies to assist radiologists in the early diagnosis of breast cancer is made possible by FV-Net.