Early Detection of Breast Cancer Subtypes with Convolutional Neural Networks
Bhagyashree Kadam, Leelkanth Dewangan, Prasanta Kumar Parida, Tanushree Chaterjee, Kirti Nahak, Ebhad Swagat Sadashiv · 2023
Breast cancer continues to be a major worldwide health issue, demanding accurate subtype categorization for personalized treatment approaches. In order to accurately identify subtypes, this work introduces a unique hybrid classifier that uses multi-modal data integration as well as cutting-edge machine learning algorithms. The model uses a Convolutional Neural Network (CNN) for image-based feature extraction alongside machine learning techniques for genomic and clinical features, combining mammographic, histopathological, genomic, and clinical data. The interpretative framework, which depends on an interpretivist philosophy, makes it less difficult to comprehend subtype discrimination on a more in-depth level. With an overall accuracy of 92.5% across HER2-positive, estrogen receptor-positive, and triple-negative subtypes, the results clearly demonstrate the hybrid classifier's better performance. The model's sensitivity and specificity ratings regularly outperformed those of existing techniques, demonstrating its aptitude for picking up on fine subtype distinctions. Particularly, the classifier demonstrated a low incidence of false positives, which is critical for reducing pointless interventions. Key genetic markers, physical traits, and hormone receptor status have been recognized as important discriminators employing feature importance analysis. These discoveries broaden the understanding of the biological foundations of each subtype. The study offers a useful tool for accurate breast cancer detection and treatment planning in clinical practice. The hybrid classifier's integrative methodology makes use of several data sources in order to provide a thorough understanding of breast cancer subtypes. This work lays the groundwork for future oncology and multi-modal data analysis studies, with potential implications in personalized therapy and other cancer types. There is still much more to discover about ethical issues, model generalization, and therapeutic use in the actual world