Contrast-enhanced transrectal ultrasound for prediction of prostate cancer aggressiveness: The role of normal peripheral zone time-intensity curves

To assess the role of time-intensity curves (TICs) of the normal peripheral zone (PZ) in the identification of biopsy-proven prostate nodules using contrast-enhanced transrectal ultrasound (CETRUS). This study included 132 patients with 134 prostate PZ nodules. Arrival time (AT), peak intensity (PI), mean transit time (MTT), area under the curve (AUC), time from peak to one half (TPH), wash in slope (WIS) and time to peak (TTP) were analyzed using multivariate linear logistic regression and receiver operating characteristic (ROC) curves to assess whether combining nodule TICs with normal PZ TICs improved the prediction of prostate cancer (PCa) aggressiveness. The PI, AUC (p < 0.001 for both), MTT and TPH (p = 0.011 and 0.040 respectively) values of the malignant nodules were significantly higher than those of the benign nodules. Incorporating the PI and AUC values (both, p < 0.001) of the normal PZ TIC, but not the MTT and TPH values (p = 0.076 and 0.159 respectively), significantly improved the AUC for prediction of malignancy (PI: 0.784–0.923; AUC: 0.758–0.891) and assessment of cancer aggressiveness (p < 0.001). Thus, all these findings indicate that incorporating normal PZ TICs with nodule TICs in CETRUS readings can improve the diagnostic accuracy for PCa and cancer aggressiveness assessment.

Scientific RepoRts | 6:38643 | DOI: 10.1038/srep38643 These quantitative indices are reported to have high diagnostic accuracy and lesser user dependency. However, the discriminative power of these indices depends in part on the "background" variability of tissue heterogeneity, which when eliminated improves the diagnostic accuracy of magnetic resonance imaging (MRI) for PCa detection 11 . However, to date, although CETRUS is a commonly used modality for the diagnosis of malignant prostate nodules, TICs of the normal peripheral zone (PZ) are still rarely taken into account.
Therefore, the aim of this study was to evaluate the role of normal PZ TICs for the detection of PCa and assessment of cancer aggressiveness, and to determine whether the diagnostic accuracy improves significantly, compared to the biopsy findings, when normal PZ quantitative parameters are incorporated.

Material and Methods
This study was approved by the ethics committee of the Jiangsu Province Hospital of Traditional Chinese Medicine, and the methods were carried out in accordance with the relevant guidelines. Written informed consent was obtained from all patients.
Patients. Between May 2014 and March 2016, 132 patients with abnormal digital rectal examination findings and/or elevated serum PSA levels (≥ 4 ng/ml and ≤ 10 ng/ml) who had not previously undergone biopsy were enrolled in this study. All the patients underwent CETRUS, which was performed by an experienced operator, and all the CETRUS data were independently reviewed frame by frame on the scanner by two other experienced investigators who were blinded to the greyscale imaging and pathological results. Then, a systematic twelve-core transrectal ultrasound-guided prostate biopsy and two-core targeted biopsy were performed based on the abnormal sonography findings by an experienced operator from the Department of Urology, who was assisted by the CETRUS operator mentioned earlier.
CETRUS procedure. Each patient was evaluated using ultrasonography at the baseline and again during intravenous infusion of sulphur hexafluoride microbubbles (SonoVue; Bracco, Milan, Italy) ( Fig. 1a,b). An IU-22 ultrasound system (Philips, Amsterdam, The Netherlands) with a transrectal end-fire transducer (c8-4v) was used. Normal greyscale imaging was performed with a probe frequency of 4 to 8 MHz and a dynamic range of 55 dB. For colour Doppler ultrasonography, the probe frequency was 6 MHz, and the colour Doppler gain was adjusted to maximize signal but eliminate colour noise from the tissue of the prostate. The colour Doppler window was set to include the entire gland. During the contrast-enhanced ultrasound examinations, a fast bolus injection of 2.4 ml SonoVue was administered intravenously; this was followed by administration of 5 ml of normal saline flush. The scanner was set in the contrast pulse-sequencing mode with a probe frequency of 8 MHz. The acoustic power was set at a mechanical index of 0.13, and the dynamic range was fixed at 55 dB. The transverse plane of the sonographic abnormality was chosen for contrast imaging. In patients with no suspicious baseline ultrasonography results, the most hypervascular plane on colour Doppler images was chosen. The entire examination was saved in a Digital Imaging and Communications in Medicine format and transmitted to a workstation for further analysis (Fig. 1c).

Image analysis.
All CETRUS data were analyzed on the workstation using the QLAB quantification software (Philips) by a sonographer who was blinded to all the clinical and pathological information (Fig. 1c). Considering that the enhancement characteristics of the PZ lesions were completely different from those of the transition zone lesions, which may reflect the hypervascularity of the normal inner gland and coexisting benign prostate hyperplasia, only the PZ was evaluated in this study. Therefore, duplicated regions of interest (ROIs) were drawn in the targeted biopsy and normal PZ site on contrast ultrasonographic images, and the diameters were set to approximately 5 mm. The TICs were reconstructed for each ROI, and then the relative quantitative parameters, which depicted the features of prostate tissue infusion in the ROI that were observed after time zero, were measured by three well-trained observers. The average of all the measurements was calculated.
Analysis of oncological outcomes. Biopsy specimens were labelled according to the location from which they were obtained and fixed with a 10% formaldehyde solution in separate test tubes. The pathological findings were assessed by an experienced pathologist as benign prostatic hyperplasia, acute or chronic prostatitis, or prostatic intraepithelial neoplasia or carcinoma (Fig. 1d). The grade of the tumour was also evaluated and assigned a standard Gleason score (Table 1).

Statistical analysis.
Student's t-test was used to analyze differences in the quantitative parameters of TICs between benign and malignant lesions ( Table 2). Multivariate logistic regression was used to test our first hypothesis: joint analysis of the nodule TICs (e.g. PI in this section) and the normal PZ PI results in better prediction of PCa. We can express the regression model for the probability of malignancy as follows: Eqs (1) and (2) are the regression equations calculated for the model using nodule PI only and the nodule and normal PZ PI together. The subscripts D and N represent the nodule and normal PZ, respectively. B and C represent the regression coefficient and regression constant, respectively, that correspond to these variables. Subsequently, the z values are calculated from Eqs (1) and (2)   from the two categorical dependent variables (benign or malignant) to calculate the probability of malignancy. However, the range of values from positive to negative is large, which makes comparison difficult. Therefore, Eq. (3) represents Poisson's conversion from z to the probability of malignancy, p, which ranged from 0 to 1. The PI D and PI N values and the significance of these variables in the multivariate logistic regression model are presented in Table 3.
Our second hypothesis was that the improved prediction of PCa results in a significant improvement in diagnostic accuracy in differentiating between benign and malignant nodules. Utilizing the receiver operating characteristic (ROC) curves constructed from malignant probability (p) values, we created a standalone PI D regression model based on Eq. (1) and a joint regression model incorporating PI N from Eq. (2) to compare diagnostic accuracy. The differences between areas under the ROC curve were calculated using statistical methods described by DeLong et al. 12 and Hanley et al. 13 Furthermore, a visual assessment of the correlation between nodule PI and normal PZ PI was provided by plotting the benign and malignant nodules with respect to their PI and the corresponding normal PZ PI.
Our third hypothesis was that including normal PZ TICs significantly improves the differentiation of cancer aggressiveness between low-grade and high-grade tumours. Generally, tumours with a Gleason score of 7, 8 or 9 are defined as high-grade tumours, whereas tumours with a Gleason score of 5 or 6 are defined as low-grade tumours 11,14 . Using box plot analysis, we established a standalone PI D regression model and incorporated the PI N model into it to compare the probability of malignancy (p) and test for significant differences.
All statistical analyses were carried out using SPSS Statistics, version 18 (IBM, Chicago, USA). P < 0.05 was considered to indicate statistical significance.

Results
Clinical and pathological characteristics. The Table 2. Total TIC analysis showed that PI (p < 0.001), MTT (p = 0.011), area under the curve (AUC) (p < 0.001) and time from peak to one half (TPH) (p = 0.040) were significantly higher in the malignant nodules than in the benign nodules, but AT, wash in slope (WIS) and TTP were not significantly higher (p = 0.512, 0.612, and 0.149, respectively; Table 2). Moreover, the TIC parameters of normal PZ tissue were not significantly different between benign and malignant lesions, except for MTT and WIS (p = 0.036 and 0.001, respectively; Table 2). Multivariate logistic regression analysis showed that nodule PI (p < 0.001), MTT (p = 0.014), AUC (p < 0.001) and TPH (p = 0.041) were significant factors with regard to the prediction of PCa, but AT, WIS and TTP were not significant factors (p = 0.509, 0.253 and 0.151, respectively; Table 3).

The effect of incorporating normal PZ PI, AUC, MTT or TPH in the prediction of PCa. Normal
PZ TICs, which reflect the "background" characteristics of prostate tissue, are correlated with the TICs of PZ in prostate nodules. By using logistic regression and Eqs (1) and (2), a regression model using only nodule TIC parameters can be expressed as Eqs (4)  In both regression models, a dramatic change was introduced by the addition of nodule PI (p < 0.001, PI D alone; p < 0.001, PI D including PI N , Table 3), nodule AUC (p < 0.001, AUC D alone; p < 0.001, AUC D including AUC N , Table 3), normal PZ PI (p < 0.001; Table 3) and AUC (p < 0.001; Table 3); this significantly improved the prediction of malignancy. Table 4 and Fig. 2a show the ROC curves for the probability value p based on the regression models of Eqs (4) and (6) Table 4 and Fig. 2b show that the area under the ROC curve based on the regression models of Eqs (5) and (7) increased by 17.5%, from 0.758 (95% CI, 0.673-0.843) to 0.891 (95% CI, 0.832-0.951) (p < 0.001, according to both the DeLong and Hanley methods).
However, incorporating the normal MTT or TPH (p = 0.285, MTT N ; p = 0.750, TPH N ; Table 3) failed to improve the prediction accuracy of PCa.
Diagnostic accuracy of the regression models. Incorporating PI N dramatically improved tumour diagnostic accuracy (p < 0.001; PI D only, p < 0.001), as did the inclusion of AUC N (p < 0.001; AUC D only, p < 0.001). When only PI D was incorporated into the regression model, the specificity, sensitivity, positive predictive value (PPV) and negative predictive value (NPV) were 73.7%, 66.7%, 64.3% and 75.7%, respectively; when PI N was also included, these values increased to 90.8% (69 of 76), 79.6% (43 of 54), 86.0% (43 of 50) and 86.3% (69 of 80), respectively. Similarly, when only AUC D was incorporated in the regression model, the specificity, sensitivity, PPV and NPV were 81.6%, 53.7%, 67.4% and 71.3%, respectively; when AUC N was also incorporated, these values   Table 4. Diagnostic performance of the nodule parameters only and both the nodule parameters and normal peripheral zone PI or AUC. PI: peak intensity, AUC: area under the curve, PPV: positive predictive value, NPV: negative predictive value. D = nodule peripheral zone tissue, N = normal peripheral zone tissue.
The AUC values are followed by the 95% CIs in parentheses.
that also incorporate PI N or AUC N (Fig. 3a,b), it seems that a relatively low PI D or AUC D might still be indicative of a highly suspicious tumour if the PI N or AUC N is also low. Thus, the diagnostic accuracy of a threshold based on only the PI D or AUC D value alone may not be very high (decision line in Fig. 3a,b).

Assessment of cancer aggressiveness based on the regression models.
Based on the probability of malignancy (p) of the regression models, it seems that both the PI D only and PI D plus PI N models are significant with regard to differentiating between low-grade tumours and benign lesions (p < 0.001 for both, Fig. 4a,b). Moreover, the PI D plus PI N model is significant for differentiating between high-grade tumours and benign lesions (p < 0.001, Fig. 4b), but the PI D only model is not (p = 0.080, Fig. 4a). Moreover, the inclusion of nodule PI along with normal PZ PI resulted in a significant improvement in the differentiation of high-grade tumours from benign nodules (p < 0.001 for both, Fig. 4a,b).

Discussion
Several studies have reported that CETRUS has high diagnostic sensitivity and specificity for the prediction of PCa 9,10 . However, with regard to TICs, the most accurate quantitative parameters differ greatly from each other 9,10 , which means that the diagnostic accuracy of the parameters vary. In this study, we found that the perfusion indices PI, MTT, AUC and TPH are significant factors with regard to the prediction of PCa (Tables 2 and 3). However, when the malignancy risk was assessed based on nodule TICs only, there was no evidence to show that any of these factors is significantly better than the others. Moreover, certain other indices that were found to be significant in previous studies on PCa 9 and liver 15,16 and breast 17 cancer did not appear significant in our findings,  for example, AT (Tables 2 and 3). It is possible that if the inter-patient variation is eliminated or heterogeneity is adjusted for, the diagnostic accuracy of these factors for PCa may improve.
In CETRUS recordings, the entire prostate is represented by groups of data points; moreover, current studies often focus on the PZ hyperperfusion region rather than the whole prostate and do not conduct tissue analysis of other normal tissue [18][19][20] . Here, we show that incorporating normal tissue TICs could significantly improve the diagnostic accuracy of nodule TICs for PCa detection (Fig. 2a,b). This is consistent with the finding that including background information enhances the discriminative ability of MRI for PCa 11 . Interestingly, the other TIC parameters, such as MTT and TPH, improved the prediction of PCa by themselves, but incorporation of the corresponding normal PZ parameters did not improve PCa prediction (Table 3). This finding indicates that "background" prostate PZ perfusion characteristics affect the "Y-axis" intensity parameters of TICs, such as PI or AUC, but they do not affect the "X-axis" time parameters such as MTT and TPH. Our results also demonstrate that low nodule TIC values are suggestive of a potential malignancy in the presence of low normal PZ TIC values. Thus, it is not feasible to set any standard diagnostic threshold values for nodules or normal parameters (Fig. 3a,b).
The incorporation of PI D plus PI N or AUC D plus AUC N significantly affected the differentiation ability of TICs. However, it not clear whether the TIC values are useful for the precise assessment of cancer aggressiveness, which is commonly estimated by the Gleason score and largely influences PCa management 21,22 . Moreover, the contrast-enhanced CETRUS findings for the PCa lesions showed various patterns according to tumour vascularity 23,24 and aggressiveness 25,26 ; this is indicative of the heterogeneity of this cancer. In our study, incorporating TICs enabled the differentiation of low-grade tumours (n = 23) from high-grade ones (n = 31) (Fig. 4b). This finding indicates that normal PZ TICs play a novel role in the assessment of cancer aggressiveness.
This study had several limitations. Firstly, the use of TICs to assess transition zone tumours has not been investigated. Given that the majority of PCa's arise in the PZ, this is a major limitation of this study. Secondly, ultrasound-guided prostate biopsy has inevitable false positive and false negative outcomes, which may have affected the final results of our study. Thirdly, the validity of the cancer aggressiveness assessment should be further tested in a large, prospective and multi-cohort study.
In conclusion, our findings demonstrate that incorporating the perfusion characteristics of normal PZ tissue may enable the identification of malignancies based on the quantitative TIC parameters of CETRUS. Thus, this is a novel approach that uses inter-patient differences for PCa prediction and assessment of cancer aggressiveness.