Diffusion Weighted Imaging Combined with Texture Analysis in Preoperative Quantitative Assessment of Rectal Cancer Staging
Objective To explore the feasibility of constructing a non-invasive model for staging rectal cancer by combining texture parameters extracted from diffusion-weighted imaging and multiphase dynamic contrast-enhanced imaging.Methods Retrospectively selected patients with rectal cancer who visited the Second Affiliated Hospital of Zhengzhou University from July 2016 to June 2021 as research subjects.All patients were staged strictly according to the 8th edition of the AJCC and were divided into an early group(Ⅰ-Ⅱ)and a late group(Ⅲ-Ⅳ).Information such as age,gender,carcinoembryonic antigen(CEA),and carbohydrate antigen 199(CA199)was collected.All patients underwent multiphase dynamic contrast-enhanced magnetic resonance(DCE-MRI)scanning and diffusion-weighted imaging before obtaining pathological materials,and vascular permeability parameters(Ktrans,Kep,Ve,Vp)as well as apparent diffusion coefficient(ADC)were calculated.Based on the DCE-MRI imaging,radiomics parameters were extracted,and a combined model was constructed using ADC,vascular permeability parameters,and clinical information for the staging prediction of rectal cancer.Results A total of 115 patients were included in this study,with 81 in the training group and 34 in the test group.Feature dimensionality reduction was performed using the radiomics feature information of patients in the training group,and the number of features corresponding to the penalty coefficient logλ=0.061 7 was used to construct the Radscore.The average Radscore value for patients in the training group stages Ⅰ-Ⅱ was-0.4(-0.7-0),and for stages Ⅲ-Ⅳ it was 0.6(0.1-1.3).For the test group,the average Radscore for stages Ⅰ-Ⅱ was-0.1(-0.5-0.2),and for stages Ⅲ-Ⅳ it was 1.2(0.5-1.4).In both the training and test groups,the Radscore of stages Ⅰ-Ⅱ was lower than that of stages Ⅲ-Ⅳ,with a statistically significant difference.After performing univariate logistic regression analysis on non-radiomics parameters for stages Ⅰ-Ⅱ and Ⅲ-Ⅳ,the OR for Ve was 0,the OR for ADC was 0.04,and the clinical model was constructed by combining them.Ultimately,a combined model was constructed using Ve,ADC,and Radscore.The AUC for the combined model in the training group was 0.921,and in the test group,it was 0.811.Conclusion The combined model constructed with clinical information,ADC,vascular permeability parameters,and radiomics parameters can assist in the non-invasive prediction of staging information for patients with rectal cancer in clinical practice.
NETL
NSTL
万方数据
