Development and Validation of a Novel Radiomics-Based Nomogram With Machine Learning to Preoperatively Predict...

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Tumor grade determines pancreatic neuroendocrine tumor (PNET) aggressiveness and guides treatment.
• Accurate preoperative prediction of PNET histology grade is needed but limited.

Purpose of the Study:

• To develop and validate a noninvasive method for predicting PNET histology grade.
• To assess the performance of a radiomic signature and a clinical-data-integrated nomogram for PNET grade prediction.

Main Methods:

• Radiomic features were extracted from three-phase CT scans (plain, arterial, venous).
• A radiomic signature was constructed using LASSO regression.
• Support Vector Machine (SVM)-linear models incorporated the radiomic signature and clinical factors to build a nomogram.

Main Results:

• A radiomic signature model stratified PNETs into grade 1 and 2/3 with AUCs of 0.911 (training) and 0.837 (validation).
• A nomogram combining plain-phase CT radiomics, T stage, and MPD/BD dilation achieved the highest performance (AUC 0.919 training, 0.875 validation).

Conclusions:

• The developed nomogram integrates radiomic signatures and clinical characteristics.
• This tool shows powerful capability for preoperative prediction of PNET grades (1 vs. 2/3).