Interpretable Deep Learning with Multi-Scale CT for Predicting Occult Lymph Node Metastasis in Early-Stage NSCLC: A...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate prediction of occult lymph node metastasis (OLNM) is vital for early-stage non-small cell lung cancer (NSCLC) treatment planning.
• Current methods for detecting OLNM can be invasive and may not always be accurate.
• Novel, non-invasive methods are needed to improve the preoperative staging of NSCLC.

Purpose of the Study:

• To develop and validate a CT-based three-dimensional (3D) deep learning model for predicting OLNM in early-stage NSCLC.
• To compare the diagnostic performance of the proposed model against other deep learning architectures and experienced radiologists.
• To assess the interpretability of the deep learning model's predictions.

Main Methods:

• A retrospective, multicenter study involving 900 patients with early-stage NSCLC.
• Development of a 3D EfficientNet deep learning model using a primary cohort (n=500) and validation on an external test cohort (n=400).
• Comparison of the model's performance against benchmark deep learning models and four experienced radiologists using AUC metrics.

Main Results:

• The 3D EfficientNet model achieved an AUC of 0.8907 in the internal test set and 0.8721 in the external test cohort.
• The model's performance was statistically superior to other convolutional neural networks and radiologists (P < 0.05).
• Interpretability analysis (Grad-CAM) revealed distinct attention patterns supporting the model's predictions.

Conclusions:

• The developed 3D EfficientNet model shows significant potential as a non-invasive tool for predicting OLNM in early-stage NSCLC.
• This AI-driven approach can enhance the accuracy of preoperative staging for lung cancer.
• The model can assist clinicians in making more precise treatment decisions for NSCLC patients.