Prediction of axillary lymph node metastasis in breast cancer using intraoperative fluorescence dual-modal imaging

In recent years, intraoperative near-infrared fluorescence imaging (FI) has been widely used in the detection of sentinel lymph nodes and lymphatic vessel imaging in breast cancer (BC). The use of FI to predict the axillary lymph node metastasis (ALNM) can help doctors reduce decision-making time and improve treatment efficiency. In this work, we primarily established a Dual-Modal Fluorescence Imaging Feature Fusion Prediction (DFI-FFP) model that integrates white light imaging (WLI), FI to predict the ALNM status. Firstly, based on the unique characteristics of various modal images, we selected distinct feature extraction networks to significantly enhance the complementarity of information across modalities. Secondly, we implemented cross-modality feature fusion leveraging the cross-attention. Additionally, a novel loss function was devised to address the issue of sample imbalance. Experimental results were quantitatively presented in terms of the area under the receiver operating characteristic curve (AUC) and accuracy (ACC). The evaluation revealed that the DFI-FFP model significantly outperformed single-modality models in predicting ALNM status. Given the current scarcity of dual-modality models specifically designed for intraoperative fluorescence data of BC lymph nodes, we compared our model with those renowned for their performance in natural image classification tasks. The experiments demonstrated that the DFI-FFP model, with its remarkable accuracy and reliability in predicting ALNM status in BC, exhibits immense potential in assisting clinical decision-making and enabling real-time ALNM diagnosis.