ABSTRACT

PURPOSE

In this study, we assessed the performance of apparent diffusion coefficient (ADC) and diffusion-weighted imaging (DWI) metrics and their ratios across different magnetic resonance imaging (MRI) acquisition settings, with or without an endorectal coil (ERC), for the evaluation of prostate cancer (PCa) aggressiveness using whole-mount specimens as a reference.

METHODS

We retrospectively reviewed the data of pCa patients with a Gleason score (GS) of 3+4 or higher who underwent prostate MRI using a 3T unit at our institution. They were divided into two groups based on the use of ERC for MRI acquisition, and patients who underwent prostate MRI with an ERC constituted the ERC (n = 55) data set, while the remaining patients accounted for the non-ERC data set (n = 41). DWI was performed with b-values of 50, 500, 1000, and 1,400 s/mm², and ADC maps were automatically calculated. Additionally, computed DWI (cDWI) was performed with a b-value of 2000 s/mm². Six ADC and two cDWI parameters were evaluated. In the ERC data set, receiver operating characteristic (ROC) curves were plotted for each metric to determine the best cutoff threshold values for differentiating GS 3+4 PCa from that with a higher GS. The performance of these cutoff values was assessed in non-ERC dataset. The diagnostic accuracies and area under the curves (AUCs) of the metrics were compared using Fisher’s exact test and De Long’s method, respectively.

RESULTS

Among all metrics, the ADC_mean-ratio yielded the highest AUC, 0.84, for differing GS 3+4 PCa from that with a higher GS. The best threshold cutoff values of ADC_mean-ratio (£0.51) for discriminating GS 3+4 PCa from that with a higher GS classified 48 patients out of 55 with an accuracy of 87.27%. However, there was no significant difference between each metric in terms of accuracy and AUC (P = 0.163 and P =214). Similarly, in the non-ERC data set, the ADC_mean-ratio provided the highest diagnostic accuracy (82.92%) by classifying 34 patients out of 41. However, Fisher’s exact test yielded no significant difference between DWI and ADC metrics in terms of diagnostic accuracy in non-ERC data (P = 0.561).

CONCLUSION

The mean ADC ratio of the tumor to the normal prostate showed the highest accuracy and AUC in differentiating GS 3+4 PCa and PCa with a higher GS across different MRI acquisition settings; however, the performance of different ADC and DWI metrics did not differ significantly.

Improvements in state-of-the-art magnetic resonance imaging (MRI) have resulted in an exponential increase in the employment of this technique for prostate cancer (PCa). Diffusion-weighted imaging (DWI) with apparent diffusion coefficient (ADC) maps is the key component of the current multiparametric MRI (mpMRI) protocol.¹ Over the years, clinical applications of mpMRI have evolved from PCa detection and staging to noninvasive characterization of tumor in which ADC metrics play a necessary role.^2-4 ADC metrics, as a preoperative noninvasive measure of the Gleason score (GS), which reflects PCa aggressiveness, have gained considerable research attention, since the preoperative GS obtained with an ultrasound-guided needle core biopsy is subject to sampling errors.⁵ Furthermore, upgrading or downgrading of GS occurs in a non-negligible proportion of the subjects when verified by a wholemount histopathological analysis after prostatectomy.⁵

Previous studies introduced several absolute ADC metrics as surrogate-markers in predicting PCa aggressiveness.^2-4 In following studies, researchers pointed out the variability of absolute ADC metrics caused by technical or patient-related factors and advocated that tumor-to-prostate ADC ratios are more reliable and better parameters,^6-9 yet several studies have failed to demonstrate any advantage of using ADC ratios.^10-12 Use of the whole prostate gland as a denominator for calculating ADC ratios, instead of healthy prostatic tissue, and the assessment of the signal intensity (SI) of PCa on DWI have been evaluated in this context with variable success.^13–15 Although ongoing discussions regarding the performance of the mentioned metrics are strongly related to the technical factors, only few studies have investigated the success of these metrics across different MRI acquisition settings.^16–18

In this study, we investigated the optimum cutoff threshold values of several quantitative ADC and DWI metrics in differentiating GS 3 + 4 PCa patients from those with higher GS using whole-mount specimens as the reference method in patients who underwent mpMRI on a 3T unit with an endorectal coil (ERC). The determined cutoff threshold values were applied to another subset of patients with PCa who underwent mpMRI on the same scanner without an ERC to test whether the ADC- and DWI-derived ratios would show better performance than their absolute counterparts.

Methods

The local ethics committee approved this retrospective study carried out between January 2016 and December 2018 and waived the need for informed consent because of the retrospective evaluation of anonymized medical data (Approval ID: 20201222). We retrospectively reviewed all consecutive patient data who underwent radical prostatectomy for PCa at our institution to identify patients with whole-mount pathology specimens yielding GS 3 + 4 or higher GS PCa.¹⁹ The inclusion criteria were as follows: (1) prostate mpMRI obtained on a 3T unit within 6 months before the operation and at least 6 weeks after the prostate biopsy to mitigate biopsy-related artifacts, (2) available serum prostate-specific antigen (PSA) levels at the time of prostate mpMRI, (3) available whole-mount specimen, and (4) index lesion with a volume >0.5 mL in a whole-mount specimen. The exclusion criteria were as follows: (1) patients who received prior androgen deprivation therapy, radiotherapy, or transurethral resection of the prostate; (2) patients with prostate mpMRI scan with incomplete sequences or inadequate image quality; and (3) index lesions with a diameter <5 mm seen on prostate mpMRI. Figure 1 illustrates the patient selection procedure of the study.

The study sample was divided into two subgroups: the ERC and the non-ERC data sets. The ERC data set comprised the patients who underwent mpMRI with an ERC and was used to determine the optimum cutoff threshold values of quantitative ADC and DWI parameters for differentiating GS 3 + 4 PCa patients from those with a higher GS. The non-ERC data set comprised the patients in whom mpMRI was performed without an ERC and was used to evaluate the diagnostic performance of the predetermined cutoff threshold values in assessing PCa aggressiveness.

All patients underwent prostate mpMRI on a 3.0 Tesla MRI scanner (Skyra, Siemens Medical Systems). Prostate mpMRI was performed with an 18-channel phased-array coil and a liquid perfluorocarbon-filled ERC (Medrad, Bayer) in the ERC data set. Prostate mpMRI was performed with the same coil in the non-ERC data set. For all examinations, butylscopolamine bromide (Buscopan, Boehringer Ingelheim) was injected to reduce bowel movements that might cause motion artifacts. The mpMRI protocol of our institution was tailored by a senior radiologist with over 26 years of prostate MRI interpretation experience considering the recommendations of the PI-RADS committee.² The prostate mpMRI protocol (in the order of the first to last technique) consisted of tri-planar T2weighted imaging, DWI, and dynamic contrast-enhanced (DCE) imaging. DWI was performed using echo-planar imaging in axial planes at different b-values of 50, 500, 1000, and 1400 s/mm². The ADC maps were calculated automatically by the software using all available b-values (Syngo via, Siemens Medical Systems) integrated into the least-square monoexponential fitting. The formula for calculating the ADC maps was as follows: ADC = −ln (S/S0)/b, where S0 is the SI of no diffusion gradients and b is the b-value. After the acquisition, computed DWI (cDWI) was calculated with a b-value of 2000 s/mm².

The detailed parameters regarding the MRI sequences are given in Supplementary Table S1. A genitourinary pathologist with 25 years of experience interpreted all prostatectomy specimens. The specimens were prepared and interpreted according to relevant international guidelines.²⁰ Macroscopic images of the specimens were digitalized. Subsequently, the pathologist highlighted the areas containing tumor foci and index lesions on the corresponding macroscopic specimen. The index tumor was accepted as the tumor focus with the highest GS. In cases involving the presence of two tumor foci with the same GS, the tumor with the larger size was accepted as the index lesion.²¹

A radiologist with 5 years of prostate mpMRI experience and the genitourinary pathologist mutually evaluated the index lesion on prostate mpMRI, with respect to the histopathological images of the whole-mount specimen to ensure a radio-pathological match. The radiologist was free to assess the radiologic reports of the patients, which were interpreted by a senior radiologist according to PI-RADS version 2. The window level for the assessment was adjusted according to the observer’s preferences for each case. The observer drew a freehand region of interest (ROI) onto the tumor using an ADC map that included the tumor with the largest diameter. Axial T2-weighted scans were reviewed as an adjunct to delineate the borders of a tumor and the prostate. They were also used as a reference to precisely identify the borders of transitional zone cancers. Care was taken to avoid placing the ROI onto the extra-prostatic tissues and tumor-free prostatic tissues. To calculate the minimum ADC value of a tumor, the observer placed several circumferential ROIs with a diameter of 1020 mm onto the tumor. Subsequently, the ROI with the lowest ADC was accepted as the minimum ADC value, as implemented in a previous study.¹⁸

In addition, the observer drew two separate ROIs for reference. First, an ROI with a minimum diameter of 10 mm onto the contralateral histologically and radiologically tumor-free prostate at the same slice in the same zone. Thereafter, a free-hand ROI covering the whole prostate at the same slice as shown in a previous study.¹⁵ Any area with scarring and inflammation was cautiously excluded; these areas commonly manifest as areas with low ADC signals while placing the ROI onto the tumor-free prostate. Moreover, the observer avoided positioning the ROI onto the prostate capsule while drawing the free-hand ROI covering the whole prostate. Subsequently, the ROIs were automatically transferred to the cDWI scans. Figures 2 and 3 show an MRI interpretation of patients with peripheral and transitional zone PCa, respectively.

The following parameters were calculated for each index lesion: the mean and minimum tumor ADC (ADC_tumor-mean and ADC_tumor-min, respectively); the tumor-free prostatic tissue ADC (ADC_normal); the mean tumor ADC ratio calculated by dividing the mean ADC of the tumor by the mean ADC of the normal contralateral tissue (ADC_mean-ratio); the minimum tumor ADC ratio calculated by dividing the minimum ADC of the tumor by the mean ADC of the normal contralateral tissue (ADC_min-ratio); the mean tumor to whole prostate ADC ratio calculated by dividing the mean ADC of the tumor by the mean ADC of the whole prostate (ADC_{mean-whole-ratio}); the minimum lesion to whole prostate ADC ratio calculated by dividing the minimum ADC of the tumor by the mean ADC of the whole prostate (ADC_{min-whole-ratio}); the mean lesion SI on cDWI (cDWI_{SI_tumor-mean}); the mean contralateral tumor-free prostatic tissue (cDWI_{SI_normal}); and the mean lesion SI ratio calculated by dividing the mean SI of the tumor by the mean SI of the contralateral normal prostate (cDWISI_mean-ratio).

Statistical analysis

Statistical analysis was performed using SPSS software version 22 (IBM). The data were presented with means and standard deviations (SD) for normally distributed continuous variables and medians (interquartile ranges, IQR) for non-normally distributed continuous variables. The categorical data were presented using frequencies and percentages. The performance of the ADC and DWI metrics in assessing PCa aggressiveness was analyzed using receiver operating characteristic (ROC) curves and area under the ROC curves (AUCs). The AUCs were presented with their standard errors (SE) and 95% CI. Youden index was used to determine the best cutoff values on the ROC curves for each metric.²² The cutoff values determined in the ERC data set were applied to the non-ERC data set, and the sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV), and accuracy were calculated. The diagnostic accuracies and AUCs of the metrics were compared using Fisher exact test and De Long method, respectively. The diagnostic metrics were presented with their 95% CI. A P value of less than .05 was considered significant.

Another radiologist with over 2 years of prostate mpMRI experience and the same pathologist jointly evaluated all the patients from the ERC data set to assess the interobserver reliability of the measurements in different sessions. Intraclass correlation coefficients were used to assess interobserver reliability. The ADC and DWI metrics showed good to excellent reliability across two readers.²³ Detailed information regarding the interobserver reliability assessment of the entire ERC data set and of each pair of measurements is given in the supplementary document (Supplementary Table S2.)

Results

Overall, 55 men with PCa were enrolled in the ERC data set (age, 63.07 ± 7.93 years; range, 43-80 years), while the non-ERC data set comprised 41 men with PCa (age, 65.1 ± 6.5 years; range, 51-78 years). The detailed clinical and pathological characteristics of the study sample are shown in Table 1. Detailed information regarding the ADC and DWI metrics of the study sample is shown in Table 2.

ADC_mean-ratio and ADC_min-ratio yielded the highest AUC, 0.84 (P = .001; 95% CI, 0.71-0.97; and SE, 6.59) and 0.83 (P = .001; 95% CI, 0.70-0.96; and SE, 6.59), respectively, for differentiating GS 3 + 4 PCa from that with a higher GS, yet the analyses did not reveal any statistically significant differences between ADC and DWI metrics in terms of AUCs (P = .214). Figure 4 shows the ROC curves of the ADC and DWI metrics. The best threshold cutoff values of ADCmean-ratio (≤0.51) and ADCmin-ratio (≤0.45) for discriminating GS 3 + 4 PCa from those with higher GS had the highest performance with an accuracy of 87.27% and 85.45%, by predicting 48 and 47 of 55 patients, respectively. However, Fisher exact test yielded no significant difference between the metrics in terms of diagnostic accuracy (P = .163). Detailed information regarding the cutoff threshold values for the ADC and DWI metrics and their diagnostic performances are shown in Table 3.

When applied to the non-ERC data set, the ADC_tumor-mean and ADC_tumor-min cutoff threshold values of ≤0.818 and ≤0.718 (× 10⁻³ mm²/s) provided an accuracy of 65.85% (27/41) and 60.97% (25/41), respectively. In the non-ERC data set, ADC_mean-ratio correctly classified 34 of 41 patients, equating a diagnostic accuracy of 82.92%. However, Fisher exact test yielded no significant difference between the DWI and ADC metrics in terms of diagnostic accuracy in the nonERC data set (P = .561). Detailed information regarding the performances of the cutoff threshold values are shown in Table 4.

Discussion

The findings of this study showed that the ADC_mean-ratio yielded the highest diagnostic performance in assessing PCa aggressiveness in the same acquisition settings with good to excellent interobserver reproducibility. However, the performance of different ADC and DWI metrics did not differ in the ERC data set that was used for determining the best cutoff values and the non-ERC data set in which the predetermined cutoff thresholds were assessed for their discriminative ability.

In a previous study, Barrett et al.⁸ examined the utility of ADC metrics in characterizing prostate tumors across a diverse set of b-values. The researchers documented that ADC ratios outperformed their absolute counterparts in predicting PCa aggressiveness. Ding et al.¹⁶ and Peng et al.¹⁷ investigated the performance of various ADC metrics in addition to several other T2weighted imaging and DCE parameters in discriminating low-grade and high-grade PCa in a study sample comprising mpMRI examinations from two different MRI scanner manufacturers. The researchers divided their study sample based on the MRI scanner and then developed several models in each data set to test in the other scanner. In both studies, the 10th percentile ADC yielded the best performance among other ADC metrics in characterizing PCa aggressiveness, yet the researchers did not investigate the performance of the ADC ratios.^{16, 17}

A recent potentially related study by Bajgiran et al.¹⁸ investigated various ADC parameters to estimate PCa aggressiveness using a mixed study sample consisting of patients who underwent mpMRI with and without ERC. The performance of the ADC_mean-ratio surpassed other metrics in discriminating high-grade prostate carcinoma in both patient groups.¹⁸ However, the researchers individually determined different cutoff threshold values for each group; they did not test these cutoff values in an independent validation data set. In this study, we have extended their research by applying cutoff threshold values of ADC and DWI metrics that were determined in the ERC data set to the patients in whom mpMRI was performed without an ERC. In concordance to the study by Bajgiran et al., the ADC_mean-ratio consistently yielded the highest diagnostic accuracy in both samples. However, in our study, no significant differences were observed between the performance of different metrics. The sample size of our study was comparably lower than that of Bajgiran et al. (218 vs. 95), and it was further reduced because of separation of the data into two different groups with regard to the use of ERC. Hence, it is likely that the lack of statistical significance between the ADC_mean-ratio and other metrics might be a consequence of the low sample size.

It should be noted that the benefitsof using an ERC are controversial. Some researchers have advocated the use of an ERC at 3T,^{24, 25} whereas others claimed that there was no increase in image quality with the application of ERC at 3T.²⁶ Nevertheless, even proponents should consider cost and patient preference; hence, using an ERC might not be suitable for every patient in daily practice. In this context, one might assume that the proportion of prostate mpMRI obtained with an ERC might be decreased in the forthcoming future as the MRI technology continues to advance, and the availability of contemporary 3T scanners steadily increases worldwide. However, currently, quantitative ADC or DWI metrics that could accurately predict PCa aggressiveness across different coil set-ups are still of clinical importance. Additionally, this study’s primary aim was not to address the best MRI set-up for assessing PCa aggressiveness; instead, it was to conceive the most stable ADC or DWI metrics across different acquisition settings. Hence, the current discussions on the advantages and disadvantages of using ERC would not invalidate the findings of the present study.

In addition to the low sample size, there are several important limitations to this study. First, the proportion of patients with transition zone cancer was low, which prevented us from conducting sub-regional analysis. Second, there was an inevitable selection bias since we only enrolled patients who underwent radical prostatectomy. Accordingly, the study cohort did not have any patients managed with active surveillance and included only a few patients with PCa with a very high GS, who rarely undergo surgery. Therefore, the findings of the present study might not be extrapolated to all patients with PCa. Third, the tumor-free ROI was selected based on both mpMRI and whole-mount histopathological images. While the former is readily available in daily practice, it is impossible to use the latter prospectively. Fourth, we employed a slice-based tumor analysis, and several previous studies have proposed that whole lesion analysis is a better and more consistent approach for analyzing PCa.¹⁵ Fifth, we did not investigate the benefits of prostate mpMRI or clinical metrics such as PI-RADS score and PSA density in assessing PCa aggressiveness or assessing whether ADC and DWI metrics add incremental diagnostic information to these metrics.^{27, 28} Nevertheless, the main aim of this study was to investigate the most stable ADC or DWI metrics across different acquisition settings rather than assessing their incremental value to such parameters. Finally, in PI-RADS version 2.1, b-values higher than 1000 s/mm²were avoided for calculating the ADC map to prevent the diffusion kurtosis effect. However, we incorporated b-values higher than 1000 s/mm², particularly 1400 s/mm², while creating ADC maps, which might negatively influence the metrics’ performance.

In conclusion, the ADC_mean-ratio consistently showed the highest diagnostic accuracy in differentiating GS 3 + 4 PCA from that with a higher GS across different data sets in which prostate mpMRI was performed with and without an ERC, though without any significant difference. Hence, we believe that the findings of the present study warrant future multi-center studies on a larger scale to precisely reveal the role of the ADC_mean-ratio as a noninvasive surrogate marker for assessing PCa aggressiveness across different imaging settings.