摘要To quantitatively analyze prostate dynamic contrast-enhanced magnetic resonance images (DCE-MRI), a retrospective study was carried out to construct a mathematical model to predict whether a lesion is cancerous or not. With known diagnosis of 52 regions-of-interest (10 malignant, 26 benign and 16 normal) in peripheral zone (PZ) of prostate from 18 patients undergoing needle biopsy and 8 normal volunteers, we combined feature parameters (k1, k2 and tp) from the linear-slope model with logistic regression to determine the statistical probability model. Prostate cancerous probability maps were derived using the model on each subject to examine the correlation between the model-based prediction and the known diagnosis. Compared the model prediction results of 52 regions-of-interest with the histopathologic results, the total agreement rate was 90.4％; the agreement rate was 60％ for the malignant and 97.6％ for the nonmalignant. The diagnostic results only referencing the probability maps by two radiologists showed a high agreement in tissue characteristics and lesion location with the combining results from other diagnostic techniques such as PSA (prostate specific antigen) level and MRS (magnetic resonance spectroscopy).