Abstract
OBJECTIVES: To explore the value of continuous-time random walk (CTRW), fractional order calculus (FROC), and stretched exponential model (SEM) in predicting for response to induction chemotherapy (IC) in nasopharyngeal carcinoma (NPC). METHODS: This prospective study included the NPC participants (n = 79) who underwent non-Gaussian (CTRW, FROC, and SEM) model from December 2023 to October 2024. Eight diffusion parameters, namely α(CTRW), β(CTRW), Dm(CTRW), β(FROC), µ(FROC), D(FROC), α(SEM), and DDC(SEM) of the primary tumor, were derived from three diffusion models before treatment. These diffusion metrics were compared between the response and non-response groups, as defined by the RECIST 1.1 criteria. Univariate and multivariate logistic analysis was used to determine the optimal diffusion metrics and clinicopathologic variables for classifying the IC response. Predictive models were established using logistic regression. Receiver operating characteristic (ROC) curves were used to evaluate their predictive ability. RESULTS: Participants enrolled in this study were classified into response group (n = 60) and non-response group (n = 19). Participants who responded well to IC had lower α(CTRW) and β(CTRW) values (p = 0.015, p = 0.011). α(CTRW) and β(CTRW) were independently associated with the response of chemotherapy in NPC (odds ratio [OR]: 0.444 [95% confidence interval [CI], 0.214-0.922], p = 0.029; 0.338 [95% CI, 0.139-0.822], p = 0.017). ROC analysis showed the predictive performance of α(CTRW), β(CTRW), and α(+)β(CTRW) values for response to IC (AUCs of 0.710, [95% CI, 0.597-0.806], 0.713 [95% CI, 0.600-0.809], and 0.829 [95% CI, 0.728-0.904], respectively) in NPC participants. CONCLUSIONS: The developed model combining α(CTRW) and β(CTRW) showed good performance in predicting treatment response to IC in NPC. RELEVANCE STATEMENT: We developed a logistic regression model based on pre-treatment non-Gaussian diffusion MRI parameters to reliably predict early response to induction chemotherapy in locally advanced nasopharyngeal carcinoma. This model may aid in personalizing treatment and minimizing unnecessary toxicity for non-responders.