Prediction of bone metastasis of prostate cancer based on intratumoral and peritumoral radiomics of MRI T2WI combined with ADC images.

OBJECTIVE

To investigate the value of intratumoral and peritumoral MRI radiomic models in predicting bone metastasis of prostate cancer patients using T2WI combined with ADC images.

MATERIALS AND METHOD

A total of 144 patients with prostate cancer who underwent preoperative MRI (T2WI and DWI) were retrospectively included. All patients were categorized into two groups based on the presence of bone metastasis. The radiomics features were calculatd for the entire tumor and 3mm-peritumoral components on pre-processed T2WI combined with ADC images. The radiomics models based on intratumoral features, peritumoral features as well as their merged features were respectively constructed. The independent risk factors of bone metastasis of prostate cancer were used to constructed clinical prediction model. The performance of the clincal model, radiomics models and clinic-imaging combined models was evaluated by the receiver operating characteristic curve and compared with the bootstrap methods. T-test was used to compare the evaluation indicators of different prediction models.

RESULTS

The clinic-imaging combined model had the best predictive efficacy among all models. The area under the curve (AUC) of the clinic-imaging combined model for predicting bone metastasis of prostate cancer in the training dataset and test dataset were 0.937 and 0.893, respectively. The accuracy, sensitivity and specificity of this model in predicting bone metastasis of prostate cancer in the training dataset were 84.2%, 91.2% and 80.6%, respectively; the accuracy, sensitivity and specificity of the testing dataset were 76.7%, 73.3% and 78.6%, respectively.

CONCLUSIONS

T2WI and ADC intratumoral and peritumoral radiomic models can be used to noninvasively predict the primary diagnosis of PCa BM, and peritumoral radiomic model can add independent predictive value. And the clinic-imaging combined model has the better predictive value.