This study aims to explore the value of PET/CT radiomics in differentiating low grade (grades 1/2) from grade 3 A of follicular lymphoma (FL) by constructing machine learning models, providing a non-invasive approach for pathological grading of FL. A total of 89 patients with pathologically confirmed FL were enrolled in this study, all of whom underwent PET/CT scans before treatment. They were grouped into two categories according to histopathological grades: those pathologically diagnosed with grade 1 and grade 2, and those pathologically diagnosed with grade 3 A. The regions of interest (ROIs) of CT and PET images were segmented by two experienced nuclear medicine physicians in the open-source and free LIFEx software, and radiomics features were extracted on the uAI platform. The dimensions of radiomics features were reduced by correlation coefficients and the least absolute shrinkage and selection operator (LASSO) regression methods, while clinical variables including PET/CT-derived parameters and clinical risk factors were selected through univariate and multivariable logistic regression. Based on the selected features, three radiomics models were constructed using logistic regression (LR), support vector machine (SVM), and random forest (RF) classifiers to differentiate low-grade from grade 3 A FL via a fivefold cross-validation strategy. Additionally, combined models using the same classifiers were built by integrating clinical features with radiomics features. The performances of such models were assessed with receiver operating characteristic (ROC) curve and decision curve analysis (DCA). In total, 2264 PET/CT radiomics features were extracted from 89 patients and 7 optimal features (including 4 of CT features and 3 of PET features) were ultimately identified for the establishment of three radiomics models. Of the three radiomics models, the LR model performed the best in the validation cohort area under the curve (AUC) = 0.858, sensitivity = 0.775, specificity = 0.755, accuracy = 0.763. The PET/CT-based radiomics had good predictive value for FL pathological classification, which can be served as a noninvasive reference tool for gold-standard biopsy to evaluate the severity of FL.
Sun et al. (Fri,) studied this question.