Texture Analysis Based on PI-RADS 4/5-Scored Magnetic Resonance Images Combined with Machine Learning to Distinguish Benign Lesions from Prostate Cancer

Background The global morbidity and mortality of prostate cancer (PCa) increase sharply every year. Early diagnosis is essential; it determines survival and outcome. So, this study extracted the texture features of apparent diffusion coefficient images in multiparametric magnetic resonance imaging (mp-MRI) and built machine learning models based on radiomics texture analysis (TA) to determine its ability to distinguish benign from PCa lesions using the Prostate Imaging Reporting and Data System (PI-RADS) 4/5 score. Methods We enrolled 103 patients who underwent mp-MRI examinations and transrectal ultrasound and magnetic resonance fusion imaging (TRUS-MRI) targeted prostate biopsy and obtained pathological confirmation at our hospital from August 2017 to January 2020. We used ImageJ software to obtain texture feature parameters based on apparent diffusion coefficient (ADC) images, then standardized texture feature parameters, and used LASSO regression to reduce multiple feature parameters; 70% of the cases were randomly selected from the PCa group and the benign prostate hyperplasia group as the training set. The remaining 30% was used as the test set. The machine learning classification model for identifying benign and malignant prostate lesions was constructed using the feature parameters after dimensionality reduction. The clinical indicators were statistically analyzed, and we constructed a machine learning classification model based on clinical indicators of benign and malignant prostate lesions. Finally, we compared the model’s performance based on radiomics texture features and clinical indicators to identify benign and malignant prostate lesions in PI-RADS 4/5 score. Results The area under the curve (AUC) of the R-logistic model test set was 0.838, higher than the R-SVM and R-AdaBoost classification models. At this time, the corresponding R-logistic classification model formula is as follow: Y_radiomics=9.396-7.464*median ADC-0.584*kurtosis+0.627*skewness+0.576*MRI lesions volume; analysis of clinical indicators shows that the corresponding C-logistic classification model formula is as follows: Y_clinical =-2.608+0.324*PSA-3.045*Fib+4.147*LDL-C, the AUC value of the model training set was 0.860, smaller than the training set R-logistic classification model AUC value of 0.936. Conclusions Radiomics combined with the machine learning classifier model has strong classification performance in identifying benign and PCa in PI-RADS 4/5 score. Various treatments and outcomes for PCa patients can be applied clinically.