Distinguishing T1-2 and T3a tumors of rectal cancer with texture analysis and functional MRI parameters.

Abstract

PURPOSE We aimed to investigate whether the texture analysis and functional magnetic resonance imaging (fMRI) could differentiate rectal cancer pathological stages T1-2 (pT1-2) and T3a (pT3a). METHODS Eighty-two rectal adenocarcinoma patients at stage pT1-2 and pT3a received T2 and fMRI examination before surgery. The latter included apparent diffusion coefficient (ADC) sequence, dynamic contrast enhancement (DCE) MRI, and intravoxel incoherent motion (IVIM) diffusion weighted imaging. Patients were grouped into early stage (pT1-2) and advanced stage (pT3a). The MRI accuracy in diagnosing rectal cancer before surgery was calculated. The differences in clinicopathological variables, quantitative parameters including ADC values, IVIM parameters (perfusion fraction [f], true diffusion coefficient [D], and pseudo- diffusion coefficient [D*]), DCE MRI parameters (transfer constant [Ktrans], reflux constant [Kep], and extravascular extracellular fractional volume [Ve]), and texture features were compared between the groups. Receiver operating characteristic (ROC) curves of texture features and fMRI parameters were generated to distinguish pT1-2 and pT3a tumors. The multivariate analysis was used to develop a predictive model and to find independent risk factors. Hosmer-Lemeshow test was used to see the fitness of the model. DeLong test was applied to compare the ROC curves of different features. Correlation of texture features and fMRI parameters with stage were calculated using r (Spearman's rank correlation coefficient). RESULTS The preoperative accuracy in differentiating pT1-2 from pT3a rectal cancer using MRI was 74.39%. Kep, Ve, and ADC showed significant differences between the groups. Kep and ADC showed negative correlation with stage. Ve correlated positively with stage. Twenty-five texture features from T2 images showed significant differences between groups, and S(0,2)SumOfSqs and WavEnLH_s_2 among these showed better performance, showing negative correlation with stage. The area under the curve (AUC) values of S(0,2)SumOfSqs, WavEnLH_s_2, ADC, Kep, and Ve were 0.721, 0.699, 0.690, 0.666, and 0.653, respectively. The multivariate analysis showed that S(0,2) SumOfSqs, WavEnLH_s_2, and ADC are risk factors for advanced tumors, and the logistic model built by Kep, Ve, S(0,2)SumOfSqs, WavEnLH_s_2, and ADC has the AUC, sensitivity, and specificity of 0.833, 88.5%, and 73.3%, respectively. ROC curve of the model showed statistical significance between S(0,2)SumOfSqs, ADC, Kep, and Ve. The P value of the Hosmer-Lemeshow test was 0.65. CONCLUSION S(0,2)SumOfSqs, WavEnLH_s_2, and ADC are risk factors for advanced rectal cancer, and the model built by Kep, Ve, S(0,2)SumOfSqs, WavEnLH_s_2, and ADC has better performance than using a single method. The application of above combinations could be beneficial to patients' accurate and individualized treatments.