Radiomics to predict PNI in ESCC.

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Perineural invasion (PNI) is a critical prognostic factor in esophageal squamous cell carcinoma (ESCC).
• Accurate preoperative assessment of PNI is essential for guiding treatment strategies and improving patient outcomes.
• Noninvasive prediction methods are needed to avoid the limitations of traditional diagnostic approaches.

Purpose of the Study:

• To evaluate the efficacy of contrast-enhanced computed tomography (CECT) based radiomics analysis in noninvasively predicting PNI in ESCC.
• To compare the performance of different machine learning models for PNI prediction using radiomics features.

Main Methods:

• Retrospective analysis of 398 ESCC patients who underwent resection.
• Extraction of 1595 radiomics features from arterial phase CECT images.
• Feature selection using ICC, Wilcoxon rank-sum test, Spearman correlation, and Boruta algorithm.
• Development and evaluation of logistic regression (LR), random forest (RF), and support vector machine (SVM) models using AUC and DCA.

Main Results:

• Six radiomics features were selected for model development.
• The random forest (RF) model demonstrated superior performance in both training (AUC=0.773) and testing (AUC=0.767) cohorts.
• RF model outperformed LR (AUC=0.627/0.638) and SVM (AUC=0.712/0.683) models.
• Decision curve analysis indicated the highest clinical utility for the RF radiomics model.

Conclusions:

• CECT-based radiomics analysis provides a noninvasive method for preoperative PNI prediction in ESCC.
• The random forest model shows significant potential for enhancing PNI prediction accuracy.
• This approach can aid in personalized treatment planning for ESCC patients.