Different Mathematical Models of Diffusion-Weighted MR Imaging in Diagnosis of Prostate Cancer
闵祥德,王良,冯朝燕,智德波,李亮,李拔森,蔡杰,邓明,可赞
DOI: https://doi.org/10.3760/cma.j.issn.1005-1201.2015.11.008
2015-01-01
Abstract:Objective To evaluate the diagnostic value of mono-exponential, bi-exponential, and stretched-exponential models of DWI in prostate cancer.Methods A retrospective study were performed in 34 patients with pathologically confirmed prostate cancer (n=18) and benign prostatic hypertrophy (n=16).Twenty eight prostate cancer ROIs, 36 stromal benign prostatic hyperplasia ROIs and 30 normal peripheral zone ROIs were analyzed.FSE T2WI, FSE T1WI and muti-b DWI were performed in all patients.ROIs were placed within the proven prostate cancer, stromal benign prostatic hyperplasia and normal peripheral zone.The parameters of mono-exponential (ADC), bi-exponential [slow diffusion coefficient (D), fast diffusion coefficient (D*) and perfusion fraction (f)] and stretched-exponential models [distributed diffusion coefficient (DDC) and stretching parameter (α)] were recorded.The mean signal intensities of every ROI at different b values were recorded in order to assess the goodness of fit for different models.The adjusted r2 was calculated to assess the goodness of fit for different models.Variance analysis and q test were used to compare the values of r2 among groups.One-way analysis of variance was used to compare the different parameters.ROC curve was performed to evaluate the diagnosis value of different parameters in prostate cancer.Results The adjusted r2 were all statistically different among three models in prostate cancer, stromal benign prostatic hyperplasia, and normal peripheral zone (all P＜ 0.05).The r2 ofbi-exponential and stretched-exponential models achieved significantly better fitting of DWI signal than the mono-exponential model (all P＜0.05).The ADC value of prostate cancer, stromal benign prostatic hyperplasia and normal peripheral zone were (0.852±0.169) × 10-3, (1.443±0.201)×10-3, (1.994±0.184) × 10-3mm2/s, respectively.The D value were (0.658 ±0.151) × 10-3, (1.149±0.171) × 10-3, (1.689±0.238) × 10-3mm2/s, respectively.DDC were (0.618±0.258) × 10-3, (1.431 ±0.329) × 10-3, (2.134±0.213) × 10-3mm2/s, respectively.α were 0.725±0.075, 0.773±0.056, 0.847±0.075, respectively.All had statistical difference (all P＜ 0.05).The area under curve (AUC) of ADC, D, DDC and α in prostate cancer diagnosis were 0.995, 0.991, 0.984, 0.773.When the cutoff were 1.100× 10-3mm2/s, 0.900 × 10-3mm2/s, 1.100 × 10-3mm2/s, and 0.727 for ADC, D, DDC, and α, the sensitivity and specificity were 92.86％, 98.48％;92.86％, 98.48％;100.00％, 92.42％;57.14％, 86.36％;respectively.Conclusions Though stretched-exponential and bi-exponential models achieved better fitting of DWI signal than the mono-exponential model, the ADC shows equal diagnosis value compared with D and DDC.The bi-exponential and stretched-exponential model can serve a supplement to mono-exponential model.