A deep learning nomogram kit for predicting metastatic lymph nodes in rectal cancer

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Preoperative diagnosis of metastatic lymph nodes (LNs) in rectal cancer is crucial for treatment planning.
• Advanced deep learning techniques like Faster Region-based Convolutional Neural Network (Faster R-CNN) have not been previously reported for this specific application.

Purpose of the Study:

• To develop and validate a deep learning-based nomogram for the preoperative prediction of metastatic lymph node status and degree in rectal cancer.
• To assess the diagnostic performance and clinical utility of the Faster R-CNN nomogram kit.

Main Methods:

• A total of 545 rectal cancer patients' MRI images were analyzed using Faster R-CNN to identify metastatic LNs.
• Multivariate regression analyses were performed to construct Faster R-CNN nomograms, which were then validated on separate datasets.
• The nomograms incorporated predictors such as age, Faster R-CNN identified metastatic LNs, and tumor differentiation.

Main Results:

• The Faster R-CNN nomogram for predicting metastatic LN status achieved areas under the curve (AUCs) of 0.862 (training) and 0.920 (validation).
• The nomogram for predicting LN metastasis degree showed AUCs of 0.859 (training) and 0.886 (validation).
• Calibration plots and decision curve analyses confirmed good calibration and clinical utility for both nomograms.

Conclusions:

• The Faster R-CNN nomogram kit demonstrates excellent performance in discrimination, calibration, and clinical utility.
• This tool provides a convenient and reliable method for the preoperative prediction of metastatic LNs in rectal cancer.