P-284 Exploratory Magnetic Resonance Imaging Histogram Biomarkers for Response Prediction to Neoadjuvant Treatment in Oesophageal/gastro-Oesophageal Cancer

The optimal modality and sequencing of neoadjuvant treatment in gastroesophageal cancer is currently being investigated in clinical trials. Early response biomarkers are needed to guide treatment adaptation. This exploratory analysis of a prospective, non-interventional clinical trial (Magnetic resonance IMaging assessment in OeSophageal cAncer) assessed the potential of histogram T2- and diffusion-weighted imaging biomarkers in response prediction to neoadjuvant treatment in oesophageal cancer. Patients with locally advanced T2-4N0-3M0 oesophageal/gastroesophageal squamous cell or adenocarcinoma underwent MRI at baseline (Timepoint_0), in week 3 (Timepoint_1) and on completion of neoadjuvant treatment (Timepoint_2). Free-hand regions of interest (ROI) were drawn on each slice with visible tumour on axial T2-weighted and axial high-b-value (b=800) diffusion-weighted MRI. Gross tumour volume (GTV), maximum tumour diameter (MTD) and 9 histogram features were extracted for the tumour at each timepoint. Univariate logistic regression analysis was used to select features with OR >2 or < 0.5 for the primary outcome of pathological response (Mandard score 1 and 2). Variables which met the cut-off criteria were used to fit multivariate logistic regression models combining clinical variables (age, gender, T-stage and N-stage) with imaging features. Performance of the combined models was compared with that of baseline clinical model and the area under the curve (AUC), and two-tailed p values were reported. 37 patients (6 female, 31 male, median 65 years, range 41-80) were included in this analysis. 13/37 patients achieved a pathological response. Diffusion-weighted MTD at Timepoint_1 (OR 0.33, 95%CI 0.14-0.81, p 0.02) and T2-weighted kurtosis at Timepoint_2 (OR 0.09, 95%CI 0.01-0.91, p 0.04) were associated with pathological response. These statistically improved the predictive performance of the clinical model (AUC 0.88 and 0.94, respectively versus 0.75 for baseline clinical model). Diffusion-weighted maximal tumour dimension at week-3 and T2-weighted kurtosis on neoadjuvant treatment completion have the potential to improve the predictive value of clinical models for response to treatment, but would require external validation in subsequent studies.