Quantitative 18F-FDG PET/CT Model for predicting pathological complete response to neoadjuvant immunochemotherapy in NSCLC: comparison with RECIST 1.1 and PERCIST.

Objective: This study aimed to evaluate the predictive value of 18F-FDG PET/CT for pathological complete response (pCR) after neoadjuvant immunochemotherapy in resectable non-small cell lung cancer (NSCLC) and develop a quantitative pCR prediction model. We compared the model's performance with RECIST 1.1 and PERCIST.

Methods: A retrospective review was conducted on patients with resectable NSCLC who received neoadjuvant immunochemotherapy from January 2020 to December 2023. Patients with both pre-treatment (18F-FDG PET/CT scan-1) and preoperative scans (18F-FDG PET/CT scan-2) were included. 18F-FDG PET/CT parameters, clinical characteristics, and follow-up data were collected. Logistic regression was used to identify independent predictors and construct the prediction model. The model's predictive performance was compared with RECIST 1.1 and PERCIST criteria. The model was validated with an external cohort from January to September 2024. Postoperative pathological results serve as the gold standard for pCR.

Results: 36 patients were included for model development, with 19 (52.8%) achieving pCR. ΔTLR% (percentage change between two scans in tumor-to-liver ratio) and SULpeak from scan-2 were significant predictors. The developed prediction model demonstrated outstanding performance with an area under the curve (AUC) of 0.975, 100% sensitivity, and 94.1% specificity. In comparison, RECIST 1.1 showed poor sensitivity (10.5%) but high specificity (100%), while PERCIST had moderate sensitivity (73.7%) and specificity (94.1%). Validation with 8 patients confirmed the model's accuracy.

Conclusions: This study suggests that 18F-FDG PET/CT, specifically the ΔTLR% and SULpeak from scan-2, is a reliable predictor of pCR in resectable NSCLC undergoing neoadjuvant immunochemotherapy. The quantitative prediction model outperforms both RECIST 1.1 and PERCIST. These findings highlight the potential clinical utility of this model, although further validation with larger cohorts is required to confirm its robustness and generalizability.