Graph-based multi-modality network for axillary lymph node metastasis prediction in early-stage breast cancer

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Accurate axillary lymph node (ALN) status evaluation is critical for early-stage breast cancer prognosis and treatment.
• Current radiomic image-based methods for ALN status classification often suffer from low diagnostic accuracy.

Purpose of the Study:

• To explore the potential of integrating clinical parameters with dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for improved ALN metastasis prediction.
• To develop a multi-modality model that provides complementary information beyond image features.

Main Methods:

• Proposed a Graph-based Multi-Modality network (GMM-Net) combining DCE-MRI and clinical parameters.
• Utilized a Text Encoder for clinical features and a Local-Global Graph Neural Module (LGGNM) for MRI features, capturing cross-region correlations.
• Employed a Multi-Modality Feature Fusion (MMF) module to integrate information from both modalities.

Main Results:

• Evaluated on a dataset of 260 breast cancer cases with 13 clinical indicators.
• GMM-Net achieved an accuracy of 0.8482 and an AUC of 0.8461.
• The proposed GMM-Net outperformed single-modality approaches in predicting ALN metastasis.

Conclusions:

• A novel graph-based multi-modality framework (GMM-Net) was developed for preoperative ALN status assessment.
• The study demonstrates GMM-Net's potential for non-invasive prediction of ALN metastasis in early-stage breast cancer.
• This approach enhances diagnostic accuracy by combining clinical and DCE-MRI data.