To evaluate the value of radiotranscriptomics analysis of gray-scale ultrasound images and contrast-enhanced ultrasound (CEUS) images in the diagnosis of benign and malignant prostate nodules.

A total of 68 patients who underwent ultrasound-guided biopsy for suspected prostate cancer (PCa) at the First Medical Center of Chinese PLA General Hospital from December 2020 to March 2022 were analyzed. All patients underwent transrectal gray-scale ultrasound and CEUS, and the images were automatically segmented and texture features were analyzed. Biopsy specimens were subjected to RNA sequencing and prostate cancer-related gene expression profiling as well as functional enrichment and pathway analysis. Random forest, Bayesian, and support vector machine (SVM) methods were used to draw the receiver operating characteristic (ROC) curve and calibration curve to evaluate the prediction efficiency of the model.

Two key texture features, ca2-GLSZM-LZHGE and GLSZM-ZSV, were obtained by radiomics. RNA sequencing identified 120 differentially expressed genes related to PCa, and the biomarkers to distinguish PCa from benign prostatic hyperplasia (BPH) were obtained by correlation analysis: ITGB3, CAV1, miR141-3p, let-7a-5p, miR25-5p, and miR200c-3p. Functional enrichment and pathway analysis identified transcriptomic alterations associated with androgen receptor status, drug resistance, proliferation, and apoptosis. The area under the ROC curve (AUC) values of the three combined dataset models (random forest, naive Bayes, and SVM) and radiomics dataset models were 0.99, 0.98, and 0.99, and 0.99, 0.95, and 0.99, respectively, which were better than those of the clinical dataset models (0.79, 0.85, and 0.92) and molecular biomarker dataset (transcriptomics) models (0.66, 0.80, and 0.86). The ROC curve and calibration curve of the combined dataset group showed that the model had good discrimination and accuracy.

Ultrasound image texture features have potential application value in the evaluation of biomarkers of PCa, and the combined radiotranscriptomics model has better predictive efficiency than the radiomics model.