Abstract
BACKGROUND: Bone metastasis of prostate cancer (PCa) is a challenging problem, leading to poor prognosis of patients. Existing biomarkers have limited sensitivity and specificity. Therefore, we urgently need a novel diagnostic tool to predict PCa bone metastasis. METHODS: Patient data and blood samples were collected according to the inclusion and exclusion criteria. logistic regression analysis was used to screen clinical indicators and miRNAs, and radiomics was used to construct a prediction model. Finally, the performance of the model was evaluated by internal verification, external verification and Delong test. Two nomograms were successfully established by analyzing clinical data, plasma miRNAs and imaging data. Nomogram 1 predicts the presence or absence of bone metastasis; Nomogram 2 predicts whether the number of bone metastases is ≥ 4. RESULTS: Nomogram 1 constructed by tPSA, hsa-miR-548o-3p and radiomics had an AUC of 0.904. The AUC of the internal training set was 0.879, the internal test set was 0.956, and the AUC of the external data set was 0.877. The calibration curve and decision curve all performed well. Nomogram 2 constructed by ALP, hsa-miR-548o-3p and radiomics had an AUC of 0.849, with an AUC of 0.916 in the internal training set, 0.806 in the internal test set and 0.839 in the external data set. The calibration curve and decision curve showed good performance. CONCLUSIONS: The combination of plasma exosomal miRNA and radiomics model has high reliability and accuracy in predicting the presence and number of bone metastases of PCa.