Development of a multi-modal learning-based lymph node metastasis prediction model for lung cancer

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Accurate lymph node metastasis prediction is crucial for non-small cell lung cancer (NSCLC) staging and treatment.
• Current methods often rely on single data modalities, potentially limiting prediction accuracy.

Purpose of the Study:

• To develop and evaluate a 3D multi-modal learning model for automated lymph node metastasis prediction in NSCLC.
• To integrate computed tomography (CT) imaging and clinical information for enhanced classification performance.

Main Methods:

• A dataset of 4239 NSCLC patients with CT images and clinical data was utilized.
• Four deep learning-based multi-modal models were constructed and assessed for lymph node classification.
• A soft-voting ensemble technique was employed to combine multiple multi-modal models for improved accuracy.

Main Results:

• Multi-modal models integrating CT images and clinical data outperformed single-modal approaches.
• The Xception multi-modal model achieved an AUC of 0.756 (internal) and 0.736 (external).
• The ensemble model (SEResNet50_DenseNet121_Xception) demonstrated superior performance with AUCs of 0.762 (internal) and 0.751 (external).

Conclusions:

• Integrating CT images and clinical information significantly enhances lymph node metastasis prediction in NSCLC.
• The 3D multi-modal model serves as a valuable auxiliary tool for evaluating lymph node metastasis in NSCLC patients.
• This model can assist in patient screening and optimize treatment planning for non-pretreated NSCLC.