The value of CT image-based texture analysis for differentiating renal primary undifferentiated pleomorphic sarcoma from three subtypes of renal cell carcinoma

Purpose: To explore the value of computed tomography (CT) image-based texture analysis for the differential diagnosis of renal primary undifferentiated pleomorphic sarcoma and three subtypes of renal cell carcinoma. Materials and methods: Eleven cases of renal primary undifferentiated pleomorphic sarcoma and 33 cases of renal cell carcinoma (including 11 cases each of clear cell carcinoma, papillary renal cell carcinoma, and chromophobe renal cell carcinoma), which were confirmed by surgical pathology, were retrospectively analyzed. All patients underwent an abdominal CT scan and a three-phase enhanced scan (except for 1 patient with undifferentiated polymorphic sarcoma without a delayed scan). MaZda software was used to manually place regions of interest (ROIs) to extract the textural features of the lesions. The texture feature selection methods included the Fisher coefficient, the joint error coefficient (POE+ACC), the mutual information (MI), and the above three methods combined (FPM). First, these 4 methods were used to select the most meaningful texture features for the identification of primary undifferentiated pleomorphic sarcoma and 3 subtypes of renal cell carcinoma. Then, raw data analysis (RDA), principal component analysis (PCA), and linear discriminant analysis (LDA) were used to identify the four types of lesions. The results are expressed as the misclassified rate (MCR). In addition, using the 4 selected categories, 2 intermediate or more senior doctors assessed and classified the imaging data of the 44 cases. Results: Based on the 4 phase CT images and the use of the Fisher, POE+ACC, MI, and FPM methods to extract texture features and RDA and PCA to identify the 44 cases, we found that the MCR was high (65.12-77.27%), and the misjudgement results showed no statistically significant differences (P>0.05). The MCR using LDA was low (6.82-59.09%) and showed a statistically significant difference from that resulting from the use of RDA and PCA (P<0.05). The lowest MCR was obtained with the FPM method using LDA and showed a statistically significant difference from those of the Fisher, POE+ACC, and MI methods. Based on the values obtained using the FPM method with LDA, no significant differences were found among the misdiagnoses in all of the phases (x2=3.526, P>0.05). The MCR was lowest in the arterial phase (6.82%, 3/44). The MCR of the imaging diagnosis was 45.45% and showed a statistically significant difference from that obtained with the FPM method using LDA (x2=17.001, P<0.05). Conclusion: The combined use of MaZda software, the FPM method to extract CT image texture features, and LDA produced the highest rate of identification of renal primary undifferentiated pleomorphic sarcoma and 3 subtypes of kidney cancer.