Radiomics Analysis of Multiparametric PET/MRI for N- and M-Staging in Patients with Primary Cervical Cancer

Area of Science:

• Oncology
• Radiology
• Medical Imaging
• Machine Learning
• Radiomics

Background:

• Accurate staging of cervical cancer is crucial for treatment planning and patient stratification.
• Multiparametric 18F-FDG PET/MR imaging offers a comprehensive approach to visualize tumor characteristics.
• Radiomics, the extraction of quantitative features from medical images, holds potential for noninvasive tumor phenotyping.

Purpose of the Study:

• To evaluate the efficacy of multiparametric 18F-FDG PET/MR imaging combined with radiomics and machine learning for predicting N-stage and M-stage in primary cervical cancer.
• To assess the performance of machine learning algorithms in distinguishing between different stages of cervical cancer based on radiomic features.

Main Methods:

• 30 patients with primary, untreated cervical cancer underwent multiparametric 18F-FDG PET/MR imaging.
• Quantitative radiomic features were extracted from segmented primary tumors using specialized software (R environment).
• Machine learning models (SVM, RBF-SVM) were trained and validated using Python (scikit-learn) for N- and M-stage prediction.

Main Results:

• The study achieved high accuracy in predicting M-stage (sensitivity 91%, specificity 92%, AUC 0.97) using SVM with SVM-RFE.
• N-stage prediction showed moderate performance (sensitivity 83%, specificity 67%, AUC 0.82) with RBF-SVM and MIFS.
• Radiomics analysis of the primary tumor alone was sufficient for predicting metastatic status.

Conclusions:

• Multiparametric PET/MRI-based radiomics analysis can effectively predict the metastatic status (M-stage) of cervical cancer.
• The predictive performance for M-stage was superior to that for N-stage.
• This approach serves as a valuable tool for noninvasive tumor phenotyping and patient stratification in cervical cancer management.