MRI-based Radiomics analysis for differentiation degree of gastric cancer

Background: Preoperative differentiation between poorly and highly differentiated gastric cancers is important for treatment decisions. Purpose: To investigate a radiomics model for preoperative differentiation between poorly and highly differentiated gastric cancers. Study type: Retrospective. Population: 239 patients with gastric cancer were included in the study. 167 patients were assigned to the training group and 72 patients comprised the testing group. Sequence: T2-weighted (T2WI), T1-weighted (T1WI), diffusion-weighted imaging(DWI)on a 3.0T MR scanner. Assessment: : A total of 2632 radiomics features were extracted from DWI and apparent diffusion coefficient (ADC) maps. Radiomics based on above features were built using four feature selection methods and three classifiers. All models were used to differentiate poorly and highly differentiated gastric cancers. Statistical tests: 1) An analysis of independent t test was performed for clinicopathological information. 2) Four feature selection methods (Least Absolute Shrinkage and Selection Operator [LASSO], Analysis Of Variance [ANOVA], Recursive Feature Elimination [RFE] and Kruskal Wallis[KW]) and three classifiers (Support Vector Machine [SVM], Linear Discriminant Analysis [LDA] and Logistic Regression [LR]) were used to construct twelve radiomics. 3) The performance of different radiomics model was assessed using area under the receiver-operating characteristic curve (AUC) and accuracy (Acc) values. Results: The model LASSO + SVM achieved the highest AUC of 0.854 with Acc of 0.793 in all radiomics model. The combination of ADC _min and radiomics achieved the highest diagnostic efficiency with AUC of 0.919 and Acc of 0.823. Data Conclusion: The combined model of radiomics and ADC _min was accurate for distinguishing poorly and highly differentiated gastric cancers.