Baseline dynamic contrast-enhanced magnetic resonance imaging derived parameters and magnetic resonance signal intensity based logistic regression model to predict response of esophageal squamous cell carcinoma to neoadjuvant immunochemotherapy

It is important to predict response of esophageal squamous cell carcinoma (ESCC) to neoadjuvant immunochemotherapy for treatment decision-making. This study aimed to explore whether dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) derived parameters and MR signal intensity can predict the response. About 82 consecutive ESCC patients undergoing pretherapeutic DCE-MRI, T2WI, and T1WI followed by neoadjuvant immunochemotherapy were prospectively enrolled, among which patients from Institution 1 were randomly stratified into training (n = 52) and internal validation (n = 15) cohorts, and those from Institution 2 were assigned to external validation cohort (n = 15). Ktrans, Kep, and Ve of ESCC and their standard deviation (SD) were generated based on DCE-MRI, mean, and SD of MR signal intensity on T1WI and T2WI were also obtained, and coefficient of variation of these parameters were calculated. In training cohort, all parameters were statistically compared between responders and non-responders. Predictive effectiveness of individual parameters with statistical difference, and the parameters based logistic regression models were evaluated using area under the receiver operating characteristic curve (AUC) in three cohorts. Mean and SD of Kep (Kep_Mean and Kep_SD, respectively) and T2WI signal intensity (T2WI_Mean and T2WI_SD, respectively) in responders were higher than in non-responders (all P-values <0.05), among which, Kep_Mean could best predict the responsiveness with AUCs of 0.858, 0.768, and 0.870; and the model (T2WI_Mean + T2WI_SD + Kep_Mean + Kep_SD) demonstrated superior predictive performance (AUCs: 0.928, 0.911, and 0.907) in training, internal and external validation cohorts, respectively. Combination of mean and SD of Kep, and T2WI signal intensity could well predict immunochemotherapy responsiveness of ESCC.