Machine learning-based radiomics analysis of PET imaging for early prediction of concurrent chemoradiotherapy...

Cervical squamous cell carcinoma is a major global health burden, with many patients presenting with locally advanced disease requiring concurrent chemoradiotherapy (CCRT). Early assessment of treatment response (TR) is critical for optimizing therapeutic strategies and improving clinical outcomes; however, conventional imaging parameters provide limited predictive value. This study aimed to develop a PET radiomics model using machine learning (ML) to predict early TR after CCRT in patients with stage II-III locally advanced cervical squamous cell carcinoma (LACSC).

This retrospective study included 184 patients with LACSC who received CCRT (2018-2021) and underwent pre-treatment ¹⁸F-FDG PET/CT within 1 week before CCRT initiation. The patients were randomly assigned to a training set (n = 128) or an internal test set (n = 56). The ROIs were delineated using ITK-SNAP. Radiomic features were extracted using Pyradiomics and selected using the least absolute shrinkage and selection operator. Five ML algorithms (random forest, logistic regression, light gradient boosting machine, extremely randomized trees, and multilayer perceptron) were evaluated in this study. The performance was assessed using the area under the curve (AUC), sensitivity, and specificity.

Eight radiomic features were identified. The random forest model performed best, with AUCs of 0.877 (95% confidence interval [CI]: 0.810-0.943) in the training set and 0.783 (95% CI: 0.648-0.918) in the test set. The sensitivity was 0.833 in both sets, and the specificity was 0.783 (training) and 0.682 (test).

In this single-center retrospective cohort study, a pre-treatment ¹⁸F-FDG PET radiomics model using random forest showed the best performance among the five evaluated ML algorithms for predicting early CCRT response (test-set AUC = 0.783). External validation is required before clinical implementation.