DWI-related texture analysis for prostate cancer: differences in correlation with histological aggressiveness and data repeatability between peripheral and transition zones

Background We investigated the correlation between texture features extracted from apparent diffusion coefficient (ADC) maps or diffusion-weighted images (DWIs), and grade group (GG) in the prostate peripheral zone (PZ) and transition zone (TZ), and assessed reliability in repeated examinations. Methods Patients underwent 3-T pelvic magnetic resonance imaging (MRI) before radical prostatectomy with repeated DWI using b-values of 0, 100, 1,000, and 1,500 s/mm2. Region of interest (ROI) for cancer was assigned to the first and second DWI acquisition separately. Texture features of ROIs were extracted from comma-separated values (CSV) data of ADC maps generated from several sets of two b-value combinations and DWIs, and correlation with GG, discrimination ability between GG of 1–2 versus 3–5, and data repeatability were evaluated in PZ and TZ. Results Forty-four patients with 49 prostate cancers met the eligibility criteria. In PZ, ADC 10% and 25% based on ADC map of two b-value combinations of 100 and 1,500 s/mm2 and 10% based on ADC map with b-value of 0 and 1,500 s/mm2 showed significant correlation with GG, acceptable discrimination ability, and good repeatability. In TZ, higher-order texture feature of busyness extracted from ADC map of 100 and 1,500 s/mm2, and high gray-level run emphasis, short-run high gray-level emphasis, and high gray-level zone emphasis from DWI with b-value of 100 s/mm2 demonstrated significant correlation, excellent discrimination ability, but moderate repeatability. Conclusions Some DWI-related features showed significant correlation with GG, acceptable to excellent discrimination ability, and moderate to good data repeatability in prostate cancer, and differed between PZ and TZ. Supplementary Information The online version contains supplementary material available at 10.1186/s41747-021-00252-y.


Key points
• Some diffusion-weighted imaging (DWI)-related texture features significantly correlated with histological aggressiveness in prostate cancer.
• Some DWI-related texture features show clinically acceptable data repeatability in prostate cancer.
• Texture features showing correlation with histological aggressiveness and repeatability differ between zones.
• DWI with b-values of 100 and 1,500 s/mm 2 may be relevant.

Background
Texture analysis of clinical imaging has been increasingly carried out to determine its correlation with histological findings, such as lesion aggressiveness and clinical outcome [1][2][3]. Texture features extracted from magnetic resonance diffusion-weighted imaging (DWI), including apparent diffusion coefficient (ADC) maps, have shown promising results. However, there is no consensus regarding the method to calculate DWI-related metrics such as monoexponential fitting, intravoxel incoherent motion, and diffusion kurtosis imaging. From a clinical perspective, ADC maps calculated from two different b-values can be simple and easy to use, but there is no consensus regarding the use of a combination of two b-values. Furthermore, there are concerns regarding the reliability of texture features which are sensitive to imaging characteristics, possibly having coincidental significance due to a larger number of parameters [4][5][6].
Magnetic resonance imaging (MRI) is a primary imaging modality used for prostate cancer. Many studies have been reported regarding the correlation between DWI-related parameters and lesion aggressiveness, such as the Gleason score (GS) and grade group (GG), with inconsistent results. It was reported that ADC entropy showed significant difference between GS of 3 + 4 and 4 + 3 but not in ADC mean [7]. Alessandrino F et al. [8] reported similar results, with no significance in ADC mean. In contrast, Itou Y et al. [9] reported that ADC median showed a significant correlation with GS and a significant difference between GS of 3+4 and 4+3. Shan Y et al. [10] reported that ADC mean showed a significant correlation with GS and a significant difference between GS of 3 + 4 and 4 + 3. Though some studies evaluated data reliability focusing on intraobserver and interobserver agreement for the same images (ADC maps) [11,12], few studies have been performed with respect to image data reliability itself. Furthermore, though the above studies dealt with cancers in the peripheral zone (PZ) and transition zone (TZ) together, Hambrock T et al. [13] reported that ADC median showed a significant correlation with Gleason grade in PZ cancer. Jyoti R et al. [14] also reported that ADC minimum was significantly correlated with GS in PZ cancer, but not in TZ cancer. These results raise the possibility that DWIrelated features may demonstrate a different relationship with tumor aggressiveness between the PZ and TZ. In a recent systematic review, Surov A et al. [15] reported that in PZ cancer, ADC moderately correlates with GS, but it weakly correlated with in TZ cancer.
This study aimed to analyze the correlation between texture features extracted from ADC maps generated from several sets of two b-value combinations or DWIs with several b-values, and GG in the PZ and TZ, separately, and to evaluate the reliability of texture features in repeated examinations.

Population, inclusion, and exclusion criteria
This study was compliant with Helsinki Declaration. The following inclusion and exclusion criteria were considered: Inclusion criteria: patients who underwent 3-T multiparametric MRI (mpMRI) at our institute, including two sets of repeated DWI acquisitions for evaluating prostate lesions with informed consent from July 2016 to May 2020. Exclusion criteria: treatment except radical prostatectomy; lesions with a longitudinal diameter < 10 mm; lesions not detected on DWI; lesions with a voxel number within the region of interest (ROI) < 50; lesions containing voxel with ADC value < 0; poor image quality. Figure 1 shows the flowchart of patient inclusion and exclusion.

MRI
MRI was performed using a 3-T system (Ingenia, Philips Healthcar, Eindhoven, The Netherlands) with a pelvic phased-array coil. No endorectal coil was used. Either 20-mg hyoscine-N-butyl-bromide or 1-mg glucagon was injected intramuscularly before examination to minimize bowel peristalsis.
A routine mpMRI protocol was applied to all patients, including sagittal, coronal, and axial T2-weighted imaging; axial DWIs; and axial dynamic contrast-enhanced imaging before and after gadolinium chelate injection of 0.1 mmol/kg gadoterate meglumine, Magnescope, Dotarem (Guerbet, Villepinte, France). For DWI, two sequential free-breathing DWI single-shot spin-echo echoplanar images were acquired. The patient remained in the same position between the two DWI acquisitions. Four b-values (0, 100, 1,000, and 1,500 s/mm 2 ) with three orthogonal diffusion probing gradients were generated. ADC maps were generated using DWIs with bvalues of 100 and 1,000 s/mm 2 , ADC map (100, 1,000) in line with the Prostate Imaging-Reporting and Data System (PI-RADS) version 2.1 (https://www.acr.org/-/ media/ACR/Files/RADS/PI-RADS/PIRADS-V2-1.pdf) for the first and second DWIs, respectively. The DWI sequence parameters are summarized in Supplemental  Table S1.

Image analysis
Image analysis including ROI assignment was performed by a consensus decision of two observers (C.T. and M.H. with 4 and over 30 years of experience in diagnostic radiology, respectively) using a Synapse Vincent 3D Image Analysis System (Fujifilm Corporation, Tokyo, Japan). For PZ cancer, the polygonal two-dimensional ROI was manually determined on the lesion in the center slice showing hyperintensity on the first DWI with a b-value of 1,500 s/mm 2 (DWI 1,500) and hypointensity on the first ADC map (100, 1,000), referring to T2weighted imaging, dynamic contrast-enhanced imaging, and whole-mount, step-sectioned histological evaluation of prostatectomy specimen. Then, the ROI was placed on the first DWI datasets of DWI 0, DWI 100, and DWI 1,000 through IVIM application of a Synapse Vincent 3D Image Analysis System. For non-peripheral transition zone (TZ) cancers, the polygonal two-dimensional ROI was manually determined on the lesion in the center slice showing hypointensity on T2-weighted images and hyperintensity on the first DWI 1,500, referring to the first ADC map (100, 1,000), dynamic contrast-enhanced imaging, and whole-mount, step-sectioned histological evaluation of prostatectomy specimen. After this, the ROI was placed on the first DWI datasets of DWI 0, DWI 100, and DWI 1,000 through intravoxel incoherent motion (IVIM) application of the Synapse Vincent 3D Image Analysis System. The same procedures were repeated for the second DWI datasets. Voxel data distributions within the ROI were rendered in comma-separated values (CSV) format (Supplemental Figs. S1 and S2) using a Synapse Vincent 3D Image Analysis System. Then, the ADC of each voxel was calculated by fitting signal intensity decay between four patterns of b-value combinations using a monoexponential curve fit: 0 and 1,000 s/mm 2 , ADC (0, 1,000); 0 and 1,500 s/mm 2 , ADC (0, 1,500); 100 and 1,000 s/mm 2 , ADC (100, 1,000); and 100 and 1,500 s/mm 2 , ADC (100, 1,500). Representative cases are shown in Figs. 2 and 3.

Results
From July 2016 to May 2020, a total of 296 patients with suspected prostate cancer underwent mpMRI including two sets of repeated DWI acquisitions for evaluating prostate lesions with informed consent. Among them, 52 patients underwent mpMRI before prostatectomy and were histologically diagnosed as prostate cancer by the institutional pathologists. There were 62 cancers with a longitudinal diameter ≥ 10 mm. Furthermore, one lesion undetected on DWI, six lesions with a voxel number within the ROI < 50, three lesions with poor image quality either in the first or second DWI, and three lesions containing voxel with ADC value < 0 were excluded.
Finally, 44 patients with 49 cancers were analyzed. The characteristics of patients and lesions are summarized in Table 1. Of them, 11 patients underwent prostate biopsy within 6 weeks (17−36 days) before mpMRI, but no clear adverse effects were included for analysis. The duration between MRI and radical prostatectomy was from 8 to 191 days (median 69 days).

Discussion
To our knowledge, this is the first study showing a difference in DWI-related texture features that demonstrate not only significant correlations with GG and discrimination ability between GG of 1 and 2, versus GG of 3, 4, and 5, but also practical data repeatability between the PZ and TZ in prostate cancer. In PZ cancer, ADC 10% based on ADC (0, 1,500) and (100, 1,500) as well as ADC 25% based on ADC (100, 1,500) satisfied moderate correlation and had acceptable discrimination and good repeatability. These results were in accordance with a systematic review reporting that ADC correlated moderately with GS (correlation coefficient of -0.48, 95% confidence interval of -0.54 to -0.42) [15]. However, Hectors SJ et al. [20] reported that SRE and LRE using bin 16 extracted from ADC map showed significance with GS. Several differences, such as analyzing the PZ and TZ together, calculating ADC with four b-values (0, 1,000, 1,600, and 2,000 s/mm 2 ), and measuring texture feature using different methods, could explain the differences. Baek T et al. [21] reported that the entropy of GLCM from ADC map generated from bvalues of 0 and 1,000 s/mm 2 showed significance with GS. The differences in analyzing the PZ and TZ together and the distribution of lesion aggressiveness (16 out of total 65 lesions were GS of 6), including 19 biopsy-proven lesions, might explain the discrepancy. When analyzed by combining PZ and TZ cancers, the entropy of GLCM based on ADC (0, 1,000) did not show significance either in bin of 8, 16, or 32 setting (Supplemental Table S2).
In TZ cancer, busyness based on ADC (100, 1,500), and HGRE, SRHGE, and HGZE based on DWI 100 demonstrated moderate correlation coefficients, excellent discrimination, and moderate data repeatability. To evaluate the effect of bin number, texture features using bin 8 and 16 were also analyzed. Similar results were obtained (Supplemental Tables S3 and S4). In general, texture features for TZ cancer tend to show higher correlation and discrimination but lower data repeatability than those for PZ cancer.
Another important finding is that ADC histogram metrics such as 10%, which showed significance in PZ cancer, showed no significance in TZ cancer (Supplemental Fig. S3). This result was not inconsistent with the results of a systematic review, which reported that ADC correlated weakly (correlation coefficient of -0.22, 95% confidence interval of -0.47 to + 0.03) with GS in TZ cancer [15]. Furthermore, ADC 10% did not show significance and some features from DWI 100 and 0 demonstrating significance in TZ cancer may indirectly support that PI-RADS 2.1 puts emphasis on the findings of T2-weighted imaging for TZ cancer, because DWI with low b-value looks similar to fat-saturated T2weighted imaging. However, it is unclear why DWIrelated features showing significance with GG differ between PZ and TZ. One possible explanation might be that while the volume of the lumen and stroma is positively correlated with ADC, that of the epithelium is negatively correlated [22], and the degree of each composition differs between the PZ and TZ [23]. This may explain the results. However, the detailed mechanism underlying this is unknown. Regarding which two b-value combination is appropriate for calculating ADC, ADC generated from DWI 100 and 1,500 would be relevant in terms of a correlation with GG (Tables 3 and 4). We cannot interpret these results with reasonable model and/or relevant hypothesis at this time but image quality improvement of DWI 1,500 due to performance advance of MRI-system would contribute to these results. In a study comparing diagnostic ability of prostate cancer based on DWI-related features, ADC value calculated from DWIs with b-values of 50 and 1,500 s/mm 2 using a mono-exponential method reported to show the highest AUC among the IVIM, kurtosis, and IVIM-kurtosis methods [24], which is consistent with ours.
Another focus of the present study is data repeatability. DWI-related features with significance for PZ cancer demonstrated good repeatability, but those for TZ cancer remained moderate. However, moderate repeatability may be acceptable in clinical practice. In a previous study, the κ value for the reproducibility of the PI-RADS 2 score in TZ was 0.525 [25]. In another study, ICCs of lesion size in the TZ were 0.80 and 0.58 for intra-reader and inter-reader analyses, respectively [26].
Texture features themselves have high potential with respect to correlation with lesion aggressiveness and clinical outcome. However, those have a tendency prone to be affected by a mild difference of the imaging data including artifacts. Therefore, reliability studies not only for observers but also for imaging data themselves should be verified sufficiently before being applied to clinical practice.
This study has some limitations. First, we analyzed patients who underwent radical prostatectomy because of the clear correlation between histology and mpMRI, but this concept would have reduced the number of the   observers, not carried out independently. We consider consensus reading would be acceptable because one of the main purposes of the present study was to assess reliability of the imaging data themselves. Finally, in both PZ and TZ cancer, the number of lesions was not sufficiently large; therefore, further analyses by combining features through logistic regression and/or discriminant analyses were not performed.
In conclusion, some DWI-related features showed significant correlation with GG and clinically acceptable data repeatability in histologically confirmed prostate cancer, and they differed between the PZ and TZ. The texture features for TZ cancer tended to show higher correlation with GG and higher discrimination ability between GG of 1 and 2 versus GG of 3, 4, and 5, but lower data repeatability than those for PZ cancer.