Prediction of Lymph Node Metastasis in Colorectal Cancer Using Intraoperative Fluorescence Multi-Modal Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate diagnosis of lymph node metastasis (LNM) is critical for colorectal cancer (CRC) treatment planning.
• Current methods like frozen sections are time-consuming and labor-intensive.
• Intraoperative fluorescence imaging combined with deep learning (DL) offers potential for improved efficiency.

Purpose of the Study:

• To develop and evaluate a multi-modal fluorescence imaging feature fusion prediction (MFI-FFP) model for enhanced LNM detection in CRC.
• To address limitations of uni-modal imaging by integrating diverse data sources.

Main Methods:

• Established an MFI-FFP model integrating white light, fluorescence, and pseudo-color lymph node imaging.
• Employed distinct feature extraction networks for each imaging modality to maximize information complementarity.
• Designed a multi-modal feature fusion (MFF) module to combine global and local features.
• Formulated a novel loss function to handle imbalanced and challenging samples.

Main Results:

• The MFI-FFP model demonstrated superior performance over uni-modal and bi-modal approaches.
• Achieved higher area under the receiver operating characteristic (ROC) curve (AUC), accuracy (ACC), and F1 score compared to existing methods.
• Outperformed other efficient image classification networks in LNM prediction.

Conclusions:

• The MFI-FFP model shows significant potential for improving intraoperative LNM diagnosis in colorectal cancer.
• This multi-modal approach offers a promising advancement in medical image analysis for cancer staging.