Objective: To discuss the correlation between histogram analysis of quantitative mono-exponential, bi-exponential and diffusion kurtosis models in diffusion weighted imaging and the Gleason score of prostate cancer, and evaluate the application value and diagnostic efficiency in identifying low and high grade prostate cancer. Methods: A total of 50 patients with histologically confirmed as prostate cancer were examined from May 2015 to May 2016 in the Second Affiliated Hospital of Soochow University using DWI performed at 3.0 T with an extended b-value range from 0 to 2 000 s/mm(2). Data were post-processed by whole tumor histogram analysis,the ROI was manually drown in DWI (b=1 000 s/mm(2)) step by step along the outline of cancer, and quantitative analysis were performed respectively by mono-exponential, bi-exponential and diffusion kurtosis models for quantification of apparent diffusion coefficients (ADCs), diffusivity D, pseudo-diffusivity D(*), perfusion fraction f, diffusion coefficients by non-Gaussian distribution (D(k)) and kurtosis coefficient (K).Then the histogram analysis was performed to get the mean, median, 25th percentile, 75th percentile, skewness and kurtosis. The correlation between histogram analysis results of these quantitative parameters and Gleason score of prostate cancer were evaluated by Spearman correlation coefficient. The diagnostic performance of histogram analysis results of each quantitative parameters in identifying low (Gleason score≤6) and high (Gleason score>6) grade prostate cancer was performed by comparing the area under the ROC curve and the curve values. Results: The values of ADC, D and D(k) (mean, median, 25th, 75th) were negatively correlated with Gleason score of prostate cancer (r value was -0.388--0.624, P<0.05). The values of D (skewness and kurtosis) had a certain correlation with Gleason score of prostate cancer (r value were 0.413 and 0.402, P<0.05). The histogram analysis results of D(*) and f had no statistically significant correlation with Gleason score of prostate cancer (P>0.05). The values of K (mean, median, 25th, kurtosis) were positively correlated with Gleason score of prostate cancer (r value was 0.423-0.699,P<0.05). The diagnostic efficiency of histogram analysis results of these quantitative parameter values in identifying low and high grade prostate cancer showed that the ADC (median), D (25th), D(k) (mean) and K (25th) had a larger area under the curve, and were 0.844, 0.873, 0.815, and 0.919 respectively, the differences of area under the curve between any two of these parameters above were not statistically significant (all P>0.05). Conclusions: The quantitative parameters of three diffusion models (ADC, D, D(k), K) in DWI are all related to the Gleason score of prostate cancer, but in the differential diagnosis of low and high grade prostate cancer, the diagnostic efficacy of mono-exponential model is sufficient. The more complex model such as bi-exponential and diffusion kurtosis may complement it in other ways.
目的： 探讨扩散加权成像（DWI）单指数、双指数和扩散峰度模型各定量参数值的直方图分析结果与前列腺癌Gleason评分的相关性，并评价其在鉴别低与中高级别前列腺癌中的诊断效能。 方法： 回顾性分析苏州大学附属第二医院2015年5月至2016年5月间经病理证实为前列腺癌的50例患者资料，男性，60~90岁。所有患者均采用3.0 T磁共振设备，行多b值DWI扫描，b值范围0~2 000 s/mm(2)。数据后处理采用全肿瘤直方图分析的方法，在扩散加权成像（DWI）（b=1 000 s/mm(2)）图上沿癌灶轮廓手动逐层勾画感兴趣区，分别采用单指数、双指数及扩散峰度模型进行定量分析，获得前列腺癌区的DWI定量参数值：表观扩散系数（ADC）值、纯水分子扩散系数D值、灌注指数f值、与灌注相关的扩散系数D(*)值、非高斯分布校正的扩散系数D(k)值和峰度系数K值，进一步对各定量参数值做直方图分析，得到其平均数、中位数、25%位数、75%位数、偏度和峰度。运用Spearman秩相关分析各定量参数值的直方图分析结果与前列腺癌Gleason评分的相关性。通过受试者工作特征（ROC）曲线及曲线下面积（AUC）比较各定量参数值的直方图分析结果鉴别低（Gleason评分≤6分）与中高（Gleason评分>6分）级别前列腺癌的诊断效能。 结果： ADC值、D值和D(k)值（平均值、中位数、25%位数、75%位数）均与前列腺癌Gleason评分间具有一定的负相关（r值为-0.388~-0.624，均P<0.05）。D值（偏度、峰度）与前列腺癌Gleason评分呈正相关（r值分别为0.413、0.402，均P<0.05）。D(*)值和f值的直方图分析结果与前列腺癌Gleason评分间的相关性差异均无统计学意义（均P>0.05）。K值（平均值、中位数、25%位数、峰度）与前列腺癌Gleason评分正相关（r值为0.423~0.699，均P<0.05）。前列腺癌区DWI定量参数值的直方图分析结果鉴别低与高级别前列腺癌的诊断效能结果表明ADC值（中位数）、D值（25%位数）、D(k)值（平均数）和K值（25%位数）具有较大的AUC，分别为0.844、0.873、0.815和0.919，其中任意两者间的差异均无统计学意义（均P>0.05）。 结论： DWI三种扩散模型定量参数（ADC、D、D(k)、K）与前列腺癌的Gleason评分均存在相关性，但在针对低与高级别前列腺癌的鉴别诊断中，单指数模型的诊断效能已足够，较为复杂的双指数模型和扩散峰度模型可能会在其他方面为其提供补充。.
Keywords: Diffusion magnetic resonance imaging; Histogram analysis; Prostatic neoplasms.