Developing a predictive model for the efficacy of neoadjuvant chemoradiotherapy in locally advanced rectal cancer using multiparametric magnetic resonance imaging: An innovative approach

A predictive model was constructed based on traditional clinical magnetic resonance imaging (MRI) data to effectively predict the efficacy of neoadjuvant chemoradiotherapy (NCRT) for locally advanced rectal cancer (LARC). A retrospective analysis was conducted on the clinicopathological and imaging data of patients who underwent MRI prior to NCRT at Nanjing Traditional Chinese Medicine Hospital between April 2022 and June 2024. A total of 149 patients with histologically confirmed LARC were included. Based on the tumor regression grade (TRG) criteria for rectal cancer, patients were classified into TRG0 (n = 31), TRG1 (n = 34), TRG2 (n = 50), and TRG3 (n = 34). All patients underwent rectal MRI before treatment, and the imaging parameters (baseline status) were extracted. A predictive model based on multiparametric MRI indicators was developed to assess and predict the efficacy of neoadjuvant therapy in LARC. Statistically significant differences (P-values < .05) were observed in the imaging parameters, including the maximum sagittal tumor diameter, apparent diffusion coefficient (ADC), number and maximum diameter of lymph nodes, node stage, extramural vascular invasion, and circumferential resection margin (CRM), before and after treatment. A prediction model was constructed based on ablation experiments, and the best predictive performance was achieved when the feature combination included the maximum sagittal tumor diameter, ADC, maximum lymph node diameter, node stage, and CRM. The average classification accuracy, area under the curve, sensitivity, and specificity reached 87.53%, 0.85, 81.26%, and 79.32%, respectively. Among the top 50 ablation experiments ranked by predictive performance, the maximum sagittal tumor diameter, ADC, and node stage had higher proportions in the feature combinations. The prediction model constructed by using the optimized attention mechanism method has a high accuracy rate in predicting the therapeutic effect of LARC based on traditional clinical imaging data.