Preoperative Prediction of Perineural Invasion in Pancreatic Ductal Adenocarcinoma Using Machine Learning Radiomics...

Area of Science:

• Oncology
• Radiology
• Data Science

Background:

• Pancreatic ductal adenocarcinoma (PDAC) poses significant diagnostic challenges.
• Accurate preoperative prediction of perineural invasion (PNI) is crucial for treatment planning in PDAC.
• Current diagnostic methods for PNI in PDAC have limitations.

Purpose of the Study:

• To evaluate the clinical utility of machine learning radiomics derived from contrast-enhanced computed tomography (CECT) for predicting PNI status in PDAC.
• To compare the performance of radiomics, clinical, and combined models in preoperative PNI prediction.

Main Methods:

• Retrospective analysis of 143 PDAC patients (100 training, 43 testing).
• Extraction and selection of radiomics features from CECT images using statistical methods and LASSO.
• Development and comparison of logistic regression, SVM, random forest, XGBoost, and decision tree models for radiomics prediction.
• Establishment of clinical models using independent predictors (e.g., CA199) and a combined model integrating clinical and radiomics features.
• Performance assessment using ROC curves and decision curve analyses (DCAs).

Main Results:

• 14 significant radiomics features were identified from 788 extracted features.
• The SVM radiomics model achieved an AUC of 0.831 in the test group.
• The clinical model (including CA199) showed an AUC of 0.644.
• The combined model demonstrated superior performance with an AUC of 0.844, accuracy of 0.767, sensitivity of 0.806, and specificity of 0.667.
• DCA confirmed the optimal clinical value of the combined model for PNI prediction.

Conclusions:

• Machine learning radiomics models can effectively predict PNI status in PDAC patients.
• The combined model integrating radiomics and clinical data significantly enhances preoperative diagnostic performance for PNI.
• This approach aids in selecting appropriate treatment strategies for PDAC.