Multimodal Fusion of Mammography Images and Genomic Biomarkers for Breast Cancer Subtype Classification

Breast cancer remains one of the leading causes of cancer-related mortality among women worldwide, necessitating accurate and early identification of molecular subtypes for personalized treatment planning. Conventional subtype classification primarily relies on histopathological and genomic analyses, which can be costly, time-consuming, and invasive. Recent advances in artificial intelligence have enabled the integration of heterogeneous biomedical data sources to improve diagnostic accuracy. This study proposes a multimodal fusion framework that combines mammography images and genomic biomarkers for breast cancer subtype classification. Mammographic features are extracted using a deep convolutional neural network (CNN), while genomic information, including key biomarkers such as HER2, ER, PR, TP53, and PIK3CA, is processed through a dedicated feature extraction network. The learned representations from both modalities are fused using a feature-level fusion strategy to capture complementary imaging and molecular characteristics associated with different breast cancer subtypes.

The proposed framework is evaluated on publicly available datasets containing mammography images and corresponding genomic profiles. Four major molecular subtypes—Luminal A, Luminal B, HER2-Enriched, and Triple-Negative Breast Cancer (TNBC)—are considered for classification. Experimental results demonstrate that multimodal fusion significantly outperforms unimodal approaches based solely on imaging or genomic data. Performance is assessed using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (ROC-AUC). The model achieves improved subtype discrimination by leveraging both phenotypic information from mammograms and genotypic information from genomic biomarkers. Furthermore, explainable AI techniques are incorporated to enhance model interpretability and facilitate clinical decision-making. The findings highlight the potential of multimodal radiogenomic approaches for advancing precision oncology and improving breast cancer diagnosis, prognosis, and treatment stratification.