Differentiating Tumor Recurrence and Pseudoprogression in Postoperative Gliomas Using Pseudo- continuous Arterial Spin Labeling (pCASL) Technique

Abstract Objective: To assess whether the pCASL technique and its radiomics features can enhance the differentiation between tumor recurrence (TR) and pseudoprogression (PsP) in postoperative glioma patients. Methods: A retrospective study of 120 postoperative glioma patients (WHO Grade 2–4) from Tongji Hospital, Wuhan, was conducted. MRI data, including T1WI, T2WI, T2FLAIR, contrast-enhanced T1WI, and pCASL, were analyzed. Final diagnoses of TR or PsP were confirmed through pathology or follow-up. Among the patients, 84 had recurrence, and 36 had PsP. Process the pCASL images to obtain the CBF parameter map, then perform N4 bias correction and normalization to obtain the rCBF parameter map. The lesion areas were outlined, and mean values for ROI were calculated. Statistical analysis included the Mann-Whitney U test and ROC curve analysis. Radiomics features were extracted from the rCBF maps. These features were then further selected and divided into training and testing sets. Machine learning models, including Support Vector Machine (SVM), logistic regression, random forest, and Gaussian Naive Bayes, were developed and subsequently validated. Results: The Mann-Whitney U test showed a significant difference in mean rCBF values between TR and PsP groups (P < 0.001). ROC analysis revealed an AUC of 0.879 (95% CI: 0.817–0.941), sensitivity of 0.846, specificity of 0.836, PPV of 0.859, and NPV of 0.821. After feature selection, seven radiomics features were retained. SVM yielded the best performance with an AUC of 0.971, sensitivity of 0.950, specificity of 0.813, PPV of 0.864, and NPV of 0.929. Conclusion: The pCASL sequence can effectively differentiate between TR and PsP in postoperative glioma patients, and combining its radiomic features can significantly improve the accuracy of discrimination.