GPDBN

The Genomic and Pathological Deep Bilinear Network (GPDBN) is a deep learning framework designed to improve the prognosis prediction of breast cancer by integrating genomic data and pathological images. Recognizing the heterogeneity of breast cancer and the limitations of existing deep learning methods that may overlook complex inter-modality and intra-modality relations, GPDBN aims to harness complementary information from both genomic and pathological modalities.

GPDBN introduces a novel approach to modeling the intricate relationships between different data types through its unique inter-modality bilinear feature encoding module. This module captures the intrinsic relationships across modalities, ensuring that the complex interactions between genomic and pathological data are fully exploited. Additionally, GPDBN incorporates two intra-modality bilinear feature encoding modules to capture relations within each modality, recognizing their importance in accurate prognosis prediction.

By combining inter-modality and intra-modality bilinear features through a multi-layer deep neural network, GPDBN leverages the complementary information between these relations for enhanced prognosis prediction.