Abstract: Background and purpose: In recent years, radiomics analysis has shown certain application value in evaluating tumor heterogeneity and predicting the early effect and prognosis after chemoradiotherapy. This study aimed to predict the pathological response after neoadjuvant chemoradiotherapy of locally advanced rectal cancer (LARC). Methods: Forty-eight patients diagnosed with T_3-4 and/or N₊ of LARC were retrospectively enrolled. All enrolled patients received baseline ¹⁸F-FDG PET/CT (PET1) before neoadjuvant chemoradiotherapy and the second ¹⁸F-FDG PET/CT (PET2) within 1 week after neoadjuvant chemoradiotherapy. The operation was performed 6-8 weeks after neoadjuvant chemoradiotherapy. PET/CT images were processed to obtain maximal standardized uptake value (SUV_max), metabolic tumor volume (MTV) and texture analysis parameters ［including the use of normalized cooccurrence matrix calculation of the entropy (entropy), contrast, and coarseness based on local gray difference roughness parameter matrix calculation (coarseness)］. Kruskal-Wallis test was used to analyze the correlation between tumor regression grade (TRG) and various parameters, and the area under curve (AUC) of the receiver operating characteristic (ROC) curve was used to evaluate the diagnostic efficacy of the parameters. The support vector machine (SVM) method was also used to analyze the enrolled cases. Results: Of all patients, 20 (41.7%) were responders and 28 (58.3%) were non-responders. The averages of contrast2 based on PET2 among responders and non-responders were 84.2±31.2 and 65.6±21.8 respectively (P=0.038), and the AUC was 0.677. The SUV_max, MTV, entropy, coarseness and their changes and contrast 1 did not show statistical significance. According to the SVM method, the sensitivities of PET1 and PET2 were 25.0% and 57.1% respectively, and the specificities were both 100.0%. Both of the predictive ratios of responders among PET1 and PET2 were 100.0%. The predictive ratios of non-responders among PET1 and PET2 were 53.9% and 66.7%, respectively. The prediction accuracy of PET1 and PET2 were 60.0% and 76.9%, respectively. Conclusion: Contrast 2 as one of the texture analysis parameters of early ¹⁸F-FDG PET/CT images and the SVM method can be used to predict the pathological response of LARC after neoadjuvant chemoradiotherapy.

Key words: Textural analysis, ¹⁸F-FDG, PET/CT, Locally advanced rectal cancer