Inflammation can complicate the ability to distinguish normal tissue from cancer in the peripheral zone of the prostate. In this work, we show that a multiparametric MRI including dynamic contrast-enhanced imaging (DCE) and diffusion-weighted imaging can distinguish inflammation in the peripheral zone of the prostate from both low-grade prostate cancer and normal tissue. A depth-restricted decision tree built on the apparent diffusion coefficient (ADC) and maximal wash-in slope on DCE correctly classified 79.6% of regions of normal tissue, inflammation, and low-grade cancer in the peripheral zone of the prostate based on pathologist-detailed regions on whole-mount resected glands.
Patients were selected from a cohort of seventy-eight patients receiving a 3T mpMRI of the prostate followed by radical prostatectomy with whole-mount resection. A licensed pathologist defined regions of healthy tissue, inflammation, and prostate cancer. Regions of Interest (ROIs) were manually transposed onto an anatomic T2-weighted sequence [FSE, FOV=180x180 mm, TR/TE=6000/95 ms, resolution=0.35x0.35x3 mm, NEX=1], and propagated to functional imaging sequences including an apparent diffusion coefficient (ADC) map from six-direction diffusion-weighted imaging [EPI, FOV=240x240 mm, TR/TE=5000/(minimum) ms, resolution=0.94x0.94x4 mm, NEX=4, b=600 s/mm2], and maps of peak enhancement, time of peak enhancement, maximal wash-in slope, and average wash-out slope created from DCE imaging [3D SPGR, FOV=440x440 mm, TR/TE=5/2 ms, resolution=1.72x1.72x2.8 mm, NEX=1, temporal resolution=10.4 seconds]. MR spectroscopic imaging (MRSI) maps [3D PRESS, TR/TE = 2000/85 ms, resolution=0.86x0.86x6 mm] were created based on the integrated area of the metabolites choline, creatine, and citrate. A trilinear interpolation was used to correct for differences in resolution during ROI propagation.
Patients whose pathologies included inflammation, low grade prostate cancer, and normal peripheral zone tissue were included in this analysis. To aid interpretation in clinical practice, and appropriately for our small sample size, a model for distinguishing inflammation was generated by a decision tree using the Gini impurity with a maximal depth of three. Post hoc Wilcoxon rank sum tests using the Holm-Sidak correction were used to characterize each feature.
Sixteen patients presented with a combined 54 regions of inflammation, low grade prostate cancer, and normal peripheral zone tissue. Signal intensities on ADC and maximal DCE wash-in slope (Figure 1) correctly classified 79.6% of regions using the decision tree model in Figure 2.
In confirmatory Wilcoxon rank sum tests, ADC in normal peripheral zone tissue (1.485±.230x10-3mm2/s) was significantly higher than that of inflammation (1.110±.241x10-3mm2/s, p=0.0002) and low grade prostate cancer (1.113±.215x10-3mm2/s, p<0.0001). However, inflammation and low grade prostate cancer were not distinguishable by ADC (p=0.950). Low grade prostate cancer had significantly faster maximal wash-in slope on DCE (1.244±.410 %baseline/minute) than inflammation (0.774±.278 %baseline/minute, p=0.0004) and normal peripheral zone tissue (0.889±.360 %baseline/minute, p=0.016). However, inflammation and normal peripheral zone tissue were not distinguishable on DCE maximal wash-in slope (p=0.569). Regions of inflammation on T2-weighted imaging, spectroscopic citrate signal, peak value and time of peak value on DCE also showed significantly different signal intensities than those of normal peripheral zone tissue (p<0.05). Similarly, inflamed regions on DCE peak value and average wash-out slope showed significantly different signal intensity than those of low grade cancer. However, these metrics were not included in our depth-restricted model to avoid overfitting.
American Cancer Society MRSG-0508701-CCE
NIH: RO1 CA148708
[1] Futterer et al, Eu Urol, 2015
[2] Rosenkrantz ete al, Am J Roentgen, 2014
[3] Desouza et al, Br J Radiol 2007
[4] PI-RADS v.2, AdMeTech Intl Prostate MRI Working Group, 2014
[5] De Marza et al, Nature Rev, 2007