Quantitative Histogram Analysis of 5.0T Multiparametric MRI for Discrimination Between Prostate Cancer and Benign Hyperplasia

Abstract

Introduction

Accurate differentiation between prostate cancer (PCa) and benign prostatic hyperplasia (BPH) remains a critical diagnostic challenge with direct implications for clinical management, in part due to the inherent subjectivity of the clinical reference standard Prostate Imaging Reporting and Data System version 2.1 (PI-RADS v2.1). This study aimed to validate the diagnostic efficacy of ultra-high-field 5.0T multiparametric magnetic resonance imaging (mpMRI) histogram analysis for PCa and BPH differentiation.

Methods

This retrospective consecutive cohort study enrolled 85 patients (41 with pathologically confirmed PCa and 44 with BPH). Fourteen standardized histogram features were extracted from apparent diffusion coefficient (ADC), intravoxel incoherent motion (IVIM) parameters (D, D*, and f), and T2 mapping sequences. A multiparametric diagnostic model was constructed and validated via 5-fold stratified cross-validation.

Results

Multiple histogram parameters showed statistically significant between-group differences after Holm-Bonferroni correction for multiple comparisons (all P < 0.001), with the minimum ADC value demonstrating strong negative correlations with serum prostate-specific antigen (PSA) level (r = –0.578, P < 0.001) and Gleason score (r = –0.767, P < 0.001); the cross-validated multiparametric model yielded a mean area under the curve (AUC) of 0.9667 (95% CI: 0.924–1.000), which achieved superior diagnostic accuracy compared with the single-parameter ADC model (P < 0.05 via DeLong test).

Conclusion

These findings suggest that 5.0T MRI-based quantitative histogram analysis is a promising noninvasive tool for differentiating PCa from BPH with high accuracy. It offers particular value for reducing diagnostic uncertainty in indeterminate PI-RADS 3 lesions and supporting personalized clinical decision-making.

Keywords

5.0T magnetic resonance imaging prostate cancer benign prostatic hyperplasia histogram analysis,prostate–specific antigen

Introduction

Prostate cancer (PCa) ranks among the leading malignant neoplasms affecting men globally, with a persistently increasing incidence rate.¹ In contrast, benign prostatic hyperplasia (BPH) represents the most prevalent nonmalignant tumor in the male population.² Despite shared clinical features such as urinary symptoms, the clinical approach to PCa and BPH differs markedly, necessitating accurate differentiation.^3-5 This distinction is vital for informing tailored treatment plans and for reducing the incidence of unwarranted medical interventions. Serum prostate–specific antigen (PSA) is the predominant biomarker for PCa screening in clinical practice^6. Nonetheless, PSA levels may be elevated in nonmalignant prostate conditions, such as prostatitis and BPH, which can precipitate an excessive number of unwarranted biopsies.^7,8

Multiparametric magnetic resonance imaging (mpMRI) has emerged as a pivotal diagnostic modality for evaluating prostate lesions, offering superior accuracy in PCa detection over conventional approaches.⁹ Advanced MRI techniques, including diffusion–weighted imaging (DWI), intravoxel incoherent motion (IVIM), and T2 mapping, provide a comprehensive assessment of tumor characteristics. Based on mpMRI, the Prostate Imaging Reporting and Data System (PI–RADS v2.1) streamlines the interpretation and documentation of mpMRI results, enhancing the precision of cancer identification and the assessment of lesion aggressiveness.¹⁰ However, without objective quantitative metrics, diagnostic accuracy remains highly reader–dependent.^11,12

A robust statistical method for texture analysis, is extensively utilized in oncology to extract a variety of image parameters. These include skewness, kurtosis, entropy, and percentile values, which collectively enhance the detailed profiling of tumor heterogeneity.¹³ Compared with conventional imaging analysis, quantitative histogram analysis offers metrics that more sensitively and accurately capture tissue microstructure complexity and heterogeneity, improving the discrimination between malignant and benign lesions. Recently, a study based on a 3.0T MRI system confirmed the high prognostic potential of quantitative histogram analysis in PCa.¹⁴

Compared with conventional 1.5T and 3.0T MRI, ultrahigh–field MRI systems (5.0T and 7.0T) significantly improved the signal–to–noise ratio and spatial resolution.^15,16 This advancement is particularly important for the detection of early–stage and subtle prostate abnormalities, which are frequently overlooked by conventional MRI. While 7.0T MRI is constrained by various physiologic considerations, 5.0T MRI has already demonstrated superior imaging performance in other abdominal organs, such as the liver,¹⁷ pancreas,¹⁸ and kidneys.¹⁹ Nevertheless, the application of 5.0T MRI in prostate imaging remains underexplored. Furthermore, no studies have specifically investigated the histogram analysis of mpMRI based on 5.0T MRI for prostate assessment.

Accordingly, this study aimed to evaluate the diagnostic performance of ultrahigh–field 5.0T multiparametric MRI combined with histogram–derived quantitative parameters for differentiating benign from malignant prostate lesions, and to investigate the associations of these quantitative imaging biomarkers with tumor aggressiveness, including PSA levels and Gleason scores.

Methods

Study Participants

This study was approved by the Institutional Review Board of Hospital. A waiver of informed consent was granted due to the retrospective nature of the study, in accordance with institutional guidelines. The research was conducted in accordance with the principles outlined in the Declaration of Helsinki (as revised in 2024).

In this retrospective, single-center observational cohort study, patients with pathologically confirmed prostate cancer (PCa) or benign prostatic hyperplasia (BPH) who underwent 5.0T MRI between January 2024 and July 2024 were consecutively enrolled. The inclusion criteria were: (1) serum PSA>4 ng/mL,²⁰ (2) absence of nonprostatic malignancies, and (3) no prior prostate–targeted therapies. The exclusion criteria were: (1) history of prior malignancies or urologically significant comorbidities, (2) prior prostate–directed therapies (surgery/radiotherapy/chemotherapy/endocrine), (3) nondiagnostic MRI quality, and (4) lesions<5 mm in diameter²¹ (Figure 1).

Figure 1.

Flow chart of patient selection

All analyses were performed on a per-patient basis. For patients with multiple prostate lesions, only the lesion with the highest PI-RADS v2.1 score was included in the analysis to avoid clustering bias.

MRI Protocol

MRI was conducted using an Ultrahigh–field 5.0T MRI scanner (uMR Jupiter; United Imaging Healthcare, Shanghai, China) equipped with a 24–channel flexible body coil and an integrated 24–channel spine coil. The MRI sequences included (1) axial T2–weighted imaging (T2WI); (2) axial T1–weighted imaging (T1WI); (3) diffusion–weighted imaging (DWI) with b values of 0, 50, 800, and 1500 s/mm²; (4) intravoxel incoherent motion (IVIM) imaging with b values ranging from 0 to 1500 s/mm²; and (5) T2 mapping with multiple echo times of 15.46 ms, 30.92 ms, 46.38 ms, 61.84 ms, 77.30 ms, 92.76 ms, and 108.22 ms. Further details on the scanning sequences and parameters are provided in Table 1.

Table 1.

Scan Parameters

Scan parameters	T2WI	T1WI	DWI	IVIM	T2 mapping
Sequence	FSE	FSE	EPI	EPI	MESE
TR(ms)	8451	624	2483	4507	1500
TE(ms)	147.9	9.98	55.1	68	15.46,30.92,46.38,61.84,
TE(ms)	147.9	9.98	55.1	68	77.30,92.76,108.22
Slice thickness(mm)	3.0	3	3.0	4.0	4.0
Spacing (mm)	0.3	0.3	0.3	0.8	2
FOV(mm²)	220×220	220×220	220×220	260×260	220×220
Matrix	320×320	320×320	144×144	160×160	256×256
B value(mm²/s)	–	–	0,50, 800,1500	0,10, 20, 30, 50,100,200,400,	–
B value(mm²/s)	–	–	0,50, 800,1500	800, 1000, 1500	–
Scan time (s)	2:07	1:53	3:04	6:30	4:44

FSE, fast spin echo; EPI, echo planar imaging; SE, spin echo; TR, repetition time; TE, echo time; FOV, field of view; DWI, diffusion–weighted imaging; IVIM, intravoxel incoherent motion.

Data Processing

Quantitative mapping, including apparent diffusion coefficient (ADC) and T2 mapping, were automatically generated using the scanner’s built-in console. The D, D*, and f were derived using the intravoxel incoherent motion (IVIM) model on a United Imaging Healthcare UWS-MR R005 workstation. All parametric(ADC, D, D*, f, and T2 mapping) were subsequently imported into 3D Slicer software (version 5.6.2) for whole-lesion quantitative histogram analysis.

Two experienced urological radiologists (A with 10 years of experience and B with 15 years of experience in prostate MRI) independently performed all measurements and were blinded to the histopathological results. The final value for each feature was calculated as the mean of the two readers’ measurements.

For histogram analysis, a manual free–hand region of interest (ROI) was delineated layer–by–layer on T2–weighted imaging (T2WI) for the PI-RADS v2.1 defined index lesion. To ensure spatial consistency, the same ROI was propagated to the corresponding ADC, D, D*, f, and T2 mapping. The ROI was carefully drawn along the visible lesion border, excluding perilesional tissue, necrotic regions, cystic cavities, calcifications, the prostatic urethra, and large perilesional vessels to minimize partial volume effects and ensure accurate lesion representation.

Prior to feature extraction, voxel intensities for each parametric map were normalized independently per patient and per sequence using Z-score standardization, to reduce inter-patient signal variability while preserving the relative distributional shape of histogram features.²² Intensity discretization was performed with a fixed number of 64 bins, consistent with the International Biomarker Standardization Initiative (IBSI) radiomics standardization guidelines, to optimize the reproducibility and discriminative power of histogram features.A total of 14 histogram features were extracted using 3D Slicer: the 10th percentile, 90th percentile, entropy, interquartile range (IQR), kurtosis, maximum, mean absolute deviation (MAD), mean, median, minimum, range, root mean square (RMS), skewness, and uniformity.

Histologic Assessment

All enrolled patients underwent transrectal ultrasound-guided biopsy or radical prostatectomy. Tissue specimens were fixed in 10% neutral buffered formalin and processed for paraffin embedding. Sections were stained with hematoxylin and eosin (H&E) and, when clinically indicated, subjected to immunohistochemical staining.To ensure diagnostic accuracy and eliminate observer bias, all histopathologic slides were independently re-evaluated by a board-certified genitourinary pathologist who was blinded to the MRI findings and clinical data. Histologic grading was performed according to the International Society of Urological Pathology (ISUP) 2019 consensus guidelines. The ISUP grading system was as follows: ISUP 1 (Gleason score ≤ 6), ISUP 2 (Gleason score 3 + 4), ISUP 3 (Gleason score 4 + 3), ISUP 4 (Gleason score 8), and ISUP 5 (Gleason score 9–10).

Statistical Analysis

Statistical analyses were conducted using SPSS (version 26.0, IBM Corp., Chicago, IL), MedCalc (version 20.1, MedCalc Software Ltd., https://www.medcalc.org), and Origin 2021 (Origin Lab Corporation, Northampton, MA, USA),and Python 3.12 (64-bit) .

Interobserver reliability for the imaging parameters was assessed using the intraclass correlation coefficient (ICC) based on a two-way random-effects model for absolute agreement. Consistency was defined as follows: ICC < 0.4 indicated poor reliability, 0.4 ≤ ICC < 0.75 indicated moderate to good reliability, and ICC ≥ 0.75 indicated excellent reliability.²³ Intraobserver reproducibility was evaluated by having Reader A repeat all ROI delineations and histogram feature extractions for all 85 patients four weeks after the initial measurement, using the same ICC model.

Normality of data distribution was tested using the Shapiro–Wilk test. For continuous data with a normal distribution, the results are presented as the mean ± standard deviation (SD), whereas nonnormally distributed data are expressed as the median and interquartile range (IQR). Categorical variables are reported as frequencies and percentages.

Group comparisons were performed using the independent samples t-test for normally distributed variables and the Mann–Whitney U test for non-normally distributed variables. To account for multiple comparisons across the 14 histogram parameters for each MRI sequence (5 sequences × 14 parameters = 70 comparisons), the Holm–Bonferroni method was applied to control the family-wise error rate at α = 0.05.Adjusted P values were used to determine statistical significance.

The diagnostic performance of significant features was evaluated using ROC curve analysis, and the area under the curve (AUC) was calculated to quantify diagnostic accuracy. The Youden index was used to determine the optimal diagnostic thresholds for each parameter. In addition, ROC analysis was performed for PI-RADS v2.1 scores as a standalone diagnostic test. The AUC of the PI-RADS model was compared with those of the histogram-based models using the DeLong test to assess the added clinical value of the histogram approach.

To construct the combined ADC-D-D*-f-T2mapping model, candidate histogram features with Holm–Bonferroni adjusted P value < 0.05 in univariate logistic regression were entered into a multivariable binary logistic regression model using the enter method.To evaluate the stability and generalizability of the combined model, a 5-fold stratified cross-validation was performed using Python 3.12 (scikit-learn library). Spearman’s rank correlation test was performed to analyze the relationships between histogram features and clinical indicators, such as PSA, as well as pathological indicators, including ISUP GG and Gleason scores. A P value of < 0.05 was considered statistically significant.

The reporting of this study conforms to STARD guidelines.²⁴

Results

Patient Characteristics

This retrospective single-center study consecutively enrolled 85 male patients between January and July 2024, including 44 patients with pathologically confirmed benign prostatic hyperplasia (BPH) and 41 with histopathologically verified prostate cancer (PCa). All analyses were performed on a per-patient basis.

Demographic and clinical baseline characteristics were well balanced between the BPH and PCa groups, with no statistically significant differences observed in age, body mass index (BMI), or smoking status (all P > 0.05). In contrast, PSA levels were significantly higher in the PCa group relative to the BPH group (median 48.30 ng/mL vs. 12.55 ng/mL; P < 0.001). Prostate Imaging Reporting and Data System version 2.1 (PI-RADS v2.1) scores also differed significantly between the two groups (P < 0.001), all BPH cases were graded 1–3,whereas 97.6% of PCa cases were graded 4–5. Full baseline clinicopathological characteristics of the study population are presented in Table 2.

Table 2.

Baseline Clinical and Pathological Characteristics of the Study Population

Characteristic	BPH (n=44)	PCa (n=41)	Test Statistic	P Value
Age (years)	73.0 (68.3–80.3)	75.0 (67.5–80.0)	Z = 0.16	0.874
BMI (kg/m²)	22.95 ± 3.13	22.67 ± 2.54	t = 0.45	0.328
Smoking status			χ² = 0.05	0.819
Smoking	14 (31.8%)	14 (34.1%)
Non-smoking	30 (68.2%)	27 (65.9%)
PSA (ng/mL)	12.55 (7.29–20.78)	48.30 (15.15–161.85)	Z = 5.02	<0.001
PI-RADS v2.1
1	11 (25.0%)	–
2	22 (50.0%)	–
3	11 (25.0%)	1 (2.4%)
4	–	18 (43.9%)
5	–	22 (53.7%)
ISUP grade group
ISUP 1(≤6)	–	9 (22.0%)
ISUP 2(3+4=7)	–	5 (12.2%)
ISUP 3(4+3=7)	–	4 (9.8%)
ISUP 4(4+4=8)	–	23 (56.1%)

Significant between‑group differences (P < 0.001).

Comparisons of Histogram Parameters

Histogram-derived parameters that effectively differentiate PCa from BPH are summarized in Table 3. Histogram analysis demonstrated that for the ADC and D sequences, the 10th percentile, 90th percentile, maximum, mean, median, minimum, and RMS values were significantly lower in PCa patients than in BPH patients (all Holm-Bonferroni corrected P < 0.001). For the D* sequence, the 10th percentile and minimum values were significantly lower in the PCa cohort, whereas the 90th percentile, IQR, maximum, MAD, mean, range, and RMS values were significantly higher relative to the BPH cohort (all corrected P < 0.001). Conversely, the f sequence demonstrated significantly lower 10th percentile, 90th percentile, maximum, mean, median, minimum, and RMS values in PCa patients compared with BPH patients. Additionally, T2 mapping analysis revealed significantly lower values for the 10th percentile, 90th percentile, entropy, IQR, maximum, MAD, mean, median, minimum, and RMS in PCa patients relative to the BPH cohort (all corrected P < 0.001).

Table 3.

Histogram Parameters of ADC, D, D*, f, and T2 Mapping in BPH and PCa Groups

Parameters	BPH (n=44)	PCa (n=41)	t/z value	P value∗
ADC (×10^-3 mm²/s)
10th percentile	0.952 ± 0.126	0.643 ± 0.136	10.852	<0.001
90th percentile	1.211 (1.121–1.301)	0.840 (0.779–0.968)	-6.666	<0.001
Entropy	3.789 ± 0.525	3.755 ± 0.500	0.303	0.952
IQR	0.118 (0.093–0.160)	0.118 (0.091–0.150)	-0.110	0.990
Kurtosis	3.201 (2.502–3.870)	3.084 (2.660–3.828)	-0.088	0.990
Maximum	1.390 ± 0.217	1.085 ± 0.199	6.744	<0.001
MAD	0.078 (0.058–0.095)	0.067 (0.056–0.093)	-0.668	0.952
Mean	1.098 (0.993–1.162)	0.734 (0.661–0.863)	-6.877	<0.001
Median	1.088 (0.984–1.156)	0.715 (0.649–0.858)	-6.785	<0.001
Minimum	0.861 (0.762–0.940)	0.552 (0.431–0.612)	-6.891	<0.001
Range	0.549 ± 0.229	0.547 ± 0.237	0.033	0.952
RMS	1.101 (0.997–1.171)	0.742 (0.665–0.867)	-6.895	<0.001
Skewness	0.357 ± 0.582	0.363 ± 0.579	-0.047	0.952
Uniformity	0.086 (0.067–0.106)	0.091 (0.069–0.107)	-0.317	0.990
D (×10^-3 mm²/s)
10th percentile	0.954 (0.825–1.053)	0.567 (0.512–0.694)	6.763	<0.001
90th percentile	1.130 (1.041–1.223)	0.733 (0.656–0.880)	6.429	<0.001
Entropy	3.012 ± 0.415	3.058 ± 0.402	-0.517	0.980
IQR	0.114 (0.082–0.130)	0.094 (0.071–0.114)	1.733	0.980
Kurtosis	2.985 (2.412–3.501)	3.012 (2.503–3.511)	-0.125	0.990
Maximum	1.200 (1.087–1.309)	0.822 (0.749–0.944)	6.051	<0.001
MAD	0.065 (0.048–0.078)	0.055 (0.045–0.066)	1.381	0.980
Mean	1.034 (0.932–1.138)	0.653 (0.581–0.791)	6.710	<0.001
Median	1.036 (0.924–1.150)	0.650 (0.572–0.791)	6.719	<0.001
Minimum	0.866 (0.766–0.986)	0.527 (0.446–0.612)	6.666	<0.001
Range	0.315 (0.262–0.415)	0.303 (0.244–0.378)	1.091	0.980
RMS	1.039 (0.936–1.144)	0.656 (0.584–0.795)	6.684	<0.001
Skewness	0.215 ± 0.301	0.208 ± 0.298	0.112	0.990
Uniformity	0.091 (0.072–0.110)	0.088 (0.069–0.107)	0.285	0.990
D* (×10^-2 mm²/s)
10th percentile	0.439 (0.389–0.464)	0.299 (0.229–0.396)	5.039	<0.001
90th percentile	1.205 (0.877–2.304)	2.802 (1.311–4.942)	-3.650	<0.001
Entropy	4.490 ± 0.740	5.000 ± 1.130	-2.460	0.208
IQR	0.317 (0.217–0.510)	0.895 (0.411–1.694)	-3.949	<0.001
Kurtosis	9.120 (5.570–17.270)	11.000 (6.280–13.260)	-0.060	1.000
Maximum	4.387 (1.653–5.430)	11.056 (3.983–19.729)	-4.002	<0.001
MAD	0.341 (0.196–0.657)	1.264 (0.574–1.991)	-4.670	<0.001
Mean	0.769 (0.651–1.062)	1.400 (0.961–2.130)	-3.826	<0.001
Median	0.600 (0.538–0.702)	0.673 (0.502–0.844)	-1.258	1.000
Minimum	0.320 (0.041–0.396)	-0.543 (-1.107–-0.001)	5.576	<0.001
Range	4.172 (1.296–5.696)	11.184 (4.341–20.836)	-4.327	<0.001
RMS	1.008 (0.746–1.499)	2.509 (1.277–3.989)	-4.239	<0.001
Skewness	2.400 (1.680–3.460)	2.690 (1.840–3.120)	-0.030	1.000
Uniformity	0.064 (0.045–0.088)	0.038 (0.021–0.061)	-4.070	<0.001
f (%)
10th percentile	21.934 (17.710–25.689)	7.800 (5.339–11.726)	6.807	<0.001
90th percentile	34.172 (29.182–38.341)	23.238 (20.674–27.121)	5.813	<0.001
Entropy	3.289 ± 0.357	3.301 ± 0.429	-0.145	0.990
IQR	6.219 (5.339–8.162)	7.086 (5.646–9.803)	-1.337	0.980
Kurtosis	3.286 (2.895–3.851)	3.280 (2.722–3.728)	0.052	0.990
Maximum	37.658 (33.339–43.971)	28.350 (25.499–32.026)	5.338	<0.001
MAD	3.924 (3.296–4.933)	4.536 (3.670–5.821)	-1.900	0.980
Mean	27.948 (23.065–31.533)	16.102 (13.128–19.650)	6.570	<0.001
Median	28.400 (22.951–32.139)	16.353 (14.065–19.615)	6.394	<0.001
Minimum	15.806 (10.350–18.516)	0.512 (-1.577–7.881)	5.541	<0.001
Range	22.292 (18.129–28.633)	24.906 (20.376–31.513)	-0.932	0.980
RMS	28.586 (23.634–32.615)	17.390 (14.275–20.157)	6.376	<0.001
Skewness	-0.217 ± 0.503	-0.214 ± 0.501	-0.028	0.990
Uniformity	0.085 (0.063–0.108)	0.082 (0.061–0.106)	0.258	0.990
T2 mapping (ms)
10th percentile	75.348 ± 12.375	62.162 ± 7.625	5.958	<0.001
90th percentile	93.587 ± 14.425	75.532 ± 9.088	6.952	<0.001
Entropy	3.882 (3.740–3.995)	3.491 (3.199–3.795)	3.531	<0.001
IQR	9.616 ± 2.503	6.816 ± 2.280	5.397	<0.001
Kurtosis	3.148 (2.651–4.468)	3.338 (2.663–4.268)	-0.220	0.990
Maximum	103.968 (99.895–112.473)	83.766 (76.007–92.128)	5.224	<0.001
MAD	5.859 ± 1.480	4.211 ± 1.310	5.444	<0.001
Mean	83.842 ± 13.057	68.607 ± 7.914	6.555	<0.001
Median	82.892 ± 12.979	68.175 ± 7.856	6.373	<0.001
Minimum	69.441 ± 11.828	56.989 ± 7.921	5.737	<0.001
Range	36.726 (28.560–42.609)	25.640 (20.700–35.000)	3.272	0.035
RMS	84.191 ± 13.085	68.835 ± 7.938	6.591	<0.001
Skewness	0.614 ± 0.501	0.439 ± 0.623	1.421	0.980
Uniformity	0.608 (0.531–0.716)	0.809 (0.635–0.940)	-3.091	0.058

Parameter names in bold indicate statistically significant between-group differences after Holm-Bonferroni correction for multiple comparisons (P value∗ < 0.05).MRI:magnetic resonance imaging; ADC: apparent diffusion coefficient; D: true diffusion coefficient; D*:pseudo-diffusion coefficient; f:perfusion fraction; IQR:interquartile range; MAD: median absolute deviation; RMS: root mean square.

Representative imaging features and whole-lesion histogram profiles of a typical PCa case are shown in Figure 2, while corresponding profiles for BPH are presented in Figure 3.

Figure 2.

Imaging findings and histogram analysis of a 59–year–old male patient diagnosed with PCa with a GG of 3 and a Gleason score of 4 + 3. (A) T2WI. (B) DWI. (C) ADC map. (D) D map. (E) f map. (F) D* map. (G) T2 mapping. (H) Histogram analysis of the ADC values. (I) Histogram analysis of D values. According to the PI–RADS v2.1 criteria, the lesion was assigned a score of 5. PCa, prostate cancer; GG, Gleason grade; T2WI, T2–weighted imaging; DWI, diffusion–weighted imaging; ADC, apparent diffusion coefficient; D, true diffusion coefficient; f, perfusion fraction; D*, pseudodiffusion coefficient

Figure 3.

Imaging findings and histogram analysis of a 70–year–old male patient with BPH. (A) T2WI. (B) DWI. (C) ADC map. (D) D map. (E) f map. (F) D* map. (G) T2 mapping. (H) Histogram analysis of the ADC values. (I) Histogram analysis of D values. According to the PI–RADS v2.1 criteria, the lesion was assigned a score of 2. BPH, benign prostatic hyperplasia

Relationships of Histogram Parameters With PSA Level and Gleason Score

Significant correlations were observed between histogram parameters and both serum PSA level and Gleason score. Among all parameters, ADC minimum showed the strongest negative correlation with PSA level(r =-0.578, P < 0.001) and and with Gleason score (r = -0.767, P < 0.001). D, f, and T2 mapping parameters also demonstrated negative correlations with these clinical and pathological indicators, whereas D* showed positive correlations. Detailed correlation results are presented in Figure 4.

Figure 4.

Correlation heatmapping of histogram–derived quantitative MRI parameters with prostate–specific antigen (PSA) levels and Gleason scores. (A) ADC, (b) D, (c) (D)*, (D) f, and (E) T2 mapping. Red indicates positive and blue negative correlations, with color intensity reflecting correlation strength

Diagnostic Performance of the PSA and Histogram Parameter Models in Differentiating PCa From BPH

ROC curve analysis based on histogram–derived parameters demonstrated variable diagnostic performance in differentiating PCa from BPH. Compared with PSA, which achieved an AUC of 0.817 (95% CI: 0.718–0.892), 70.73% sensitivity, and 81.82% specificity. For PI-RADS v2.1 as a standalone diagnostic test, the AUC was 0.997 (95% CI: 0.990–1.000),97.56% sensitivity, and 100% specificity.Among single–parameter models, the ADC model yielded the highest accuracy, with an AUC of 0.948 (95% CI: 0.878–0.985), 92.68% sensitivity, and 88.64% specificity. The D model also performed robustly, with an AUC of 0.940 (95% CI: 0.866–0.980), 87.80% sensitivity, and the highest specificity of 93.18%. The f model demonstrated comparable efficacy (AUC 0.935, 95% CI: 0.872–0.977) with balanced sensitivity (87.80%) and specificity (90.91%). By contrast, the D* and T2 mapping models showed relatively lower diagnostic accuracy, with AUC of 0.908 and 0.897, respectively. Notably, the combined multiparametric model integrating ADC, D, D*, f, and T2 mapping achieved the best overall diagnostic performance, with an AUC of 0.975 (95% CI: 0.935–0.993), 95.12% sensitivity, and 95.45% specificity. The detailed results are presented in Table 4 and Figure 5.

Table 4.

Comparison of the Diagnostic Efficiency in Seven Models

Model	AUC (95%CI)	P value	Sensitivity (%)	Specificity (%)
PSA	0.817 (0.718–0.892)	0.001	70.73	81.82
PI-RADS v2.1	0.997 (0.990–1.000)	0.001	97.56	100
ADC model	0.948 (0.878–0.985)	0.005	92.68	88.64
D model	0.940 (0.866–0.980)	0.013	87.80	93.18
D* model	0.908 (0.855–0.975)	0.079	87.80	81.82
f model	0.935 (0.872–0.977)	0.004	87.80	90.91
T2 mapping model	0.897 (0.831–0.960)	0.145	82.93	81.82
ADC_D_D*_f_T2 mapping	0.975 (0.935–0.993)	0.001	95.12	95.45

Figure 5.

ROC curves of the PSA, ADC, D, D*, f, T2 mapping, and combined ADC_D_D*_f_T2 mapping models for differentiating PCa from BPH

The 5-fold stratified cross-validation (stratified by the pathological diagnosis of PCa and BPH) was performed to verify the generalizability and stability of the multiparametric MRI histogram-based combined diagnostic model, with 68 cases in the training set and 17 cases in the test set for each fold. The test set AUC of each fold was 0.9722, 0.9306, 1.0000, 0.9306, and 1.0000, respectively. The mean test AUC across 5 folds was 0.9667, with a SD of 0.035, and the 95% confidence interval (CI) of the mean AUC was 0.924–1.000. To adjust for the potential overestimation of diagnostic performance in the full-cohort model (full-cohort AUC: 0.975), optimism correction was conducted using the cross-validation-based bias correction method. The optimism value, defined as the absolute difference between the full-cohort AUC and the 5-fold mean test AUC, was 0.018, with a final optimism-corrected AUC of 0.9667. The detailed cross-validation results are presented in Supplementary Table S2.

Interobserver Reproducibility

The interobserver consistency analysis, which used the ICC, demonstrated a high level of agreement for all evaluated histogram parameters (ICC > 0.75). The specific ICC values were as follows: ADC (0.949), D (0.909), f (0.829), D* (0.870), and T2 mapping (0.848). The detailed results of the interobserver consistency analysis are presented in Table 5.

Table 5.

Interobserver Reproducibility

Mean	Observer 1	Observer 2	ICC	95%CI
ADC(x103mm2/s)	0.939 ± 0.218	0.933 ± 0.218	0.949	0.917–0.969
D(x10–3mm2/s)	0.805 ± 0.176	0.796 ± 0.175	0.909	0.854–0.944
F	0.240 ± 0.86	0.234 ± 0.078	0.829	0.717–0.896
D*(x103mm2/s)	1.604 ± 0.690	1.488 ± 0.712	0.870	0.784–0.922
T2mapping(ms)	76.208 ± 13.840	74.680 ± 15.296	0.848	0.750–0.908

Intraobserver Reproducibility

Intraobserver reproducibility was excellent for all parameters, with ICC values exceeding 0.98 for all sequences. The highest reproducibility was observed for the D sequence (mean ICC = 0.995), followed by f (0.994), ADC (0.991), D* (0.989), and T2 mapping (0.986). Notably, several features demonstrated very high ICC values, including D entropy (0.999), f10th percentile (0.999), and ADC median (0.999).Detailed results are provided in Supplementary Table S1.

Discussion

This study employed multiparametric histogram analysis on ultra-high-field 5.0T MRI to evaluate its diagnostic performance in distinguishing PCa from BPH. Significant differences were observed in histogram-derived parameters from ADC, D, D*, f, and T2 mapping between PCa and BPH patients. Furthermore, these histogram parameters showed mild-to-moderate correlations with PSA and Gleason scores. These findings suggest that 5.0T MRI-based multiparametric histogram parameters may predict prognostic characteristics of PCa and could enhance risk assessment accuracy.

Histogram Analysis of ADC, D, D*, f, and T2 Mapping Parameters for Differentiating PCa From BPH

Histogram analysis demonstrated significant differences in multiparametric features (ADC, D, D*, f, and T2 mapping) between PCa and BPH lesions, validating the diagnostic utility of these metrics at 5.0T, consistent with previous reports based on 3.0T MRI.²⁵ Notably, T2 mapping values in PCa were markedly lower than those in BPH, and the tissue contrast between malignant and benign lesions was further amplified at 5.0T relative to 3.0T, likely attributable to field strength-dependent relaxation effects.²⁶ To our knowledge, this is the first study to validate these multiparametric histogram differences at 5.0T, underscoring its potential as a robust imaging platform for distinguishing PCa from BPH.

Correlation Between Histogram Parameters and Clinical and Pathological Indicators in Prostate Cancer

Our study revealed significant correlations between histogram-derived parameters, clinical indicators (PSA levels), and pathological measures (ISUP grade group, Gleason score). The minimum ADC value demonstrated the strongest negative correlation with PSA levels and Gleason scores, aligning with previous evidence.²⁷ This underscores that restricted water diffusion reflects increased cellular density and serves as an imaging marker of tumor aggressiveness in PCa. Consistent with 3.0T MRI observations,²⁸ these findings support the role of 5.0T-derived histogram parameters as noninvasive biomarkers for PCa risk stratification and personalized management.

Diagnostic Efficacy for Differentiating PCa From BPH

Our study showed that histogram-derived parameters from ADC, D, f, and D* had good discriminatory value in differentiating pathologically confirmed PCa from BPH, with lower-percentile metrics showing particular diagnostic relevance, and all parameters showing significant between-group differences after Holm-Bonferroni correction (all adjusted P<0.05). Additionally, the combined ADC_D_D*_f_T2 mapping model achieved an excellent full-cohort AUC of 0.975, with both sensitivity and specificity exceeding 95%, and 5-fold stratified cross-validation further supported the robust stability and generalizability of the model, with a mean test AUC of 0.9667 (95% CI: 0.924–1.000) and a minimal optimism value of 0.008, indicating negligible overfitting. The combined model appeared superior to single-parameter approaches, as confirmed by DeLong test (all P<0.05). By integrating metrics of water diffusion (ADC and D), tissue perfusion (D and f), and T2 relaxation properties, the multiparametric model may provide a more comprehensive characterization of tumor biology and microstructural heterogeneity, potentially enhancing the accuracy of distinguishing PCa from BPH.

Diagnostic Value of 5.0T MRI in Prostate Disease Management

Multiple recent prospective comparative studies have consistently demonstrated that 5.0T MRI provides significantly improved image quality, SNR, and lesion conspicuity compared to 3.0T MRI for prostate imaging, without a corresponding increase in clinically significant artifacts.^29-31 Given these technical advantages, our study further shows that histogram-derived quantitative parameters from 5.0T mpMRI achieve high diagnostic accuracy for differentiating PCa from BPH. A prior meta-analysis reported pooled AUC values of 0.87 (ADC), 0.85 (D), 0.75 (f), and 0.76 (D*) in identifying prostate cancer.³² In the present investigation, 5.0T MRI yielded AUC values of 0.948 (ADC), 0.940 (D), 0.908 (D*), and 0.935 (f), all numerically higher than those reported in analogous 3.0T histogram studies. This improved performance may be partly attributable to increased SNR and CNR, as well as optimized acquisition protocols, advanced coil design, and a favorable balance between SNR enhancement and artifact mitigation at ultra-high-field strength.

Notably, our combined model performed comparably to the clinical standard PI-RADS v2.1 (DeLong test, P= 0.212). While PI-RADS v2.1 showed excellent diagnostic performance, our histogram-based model offers objective, quantitative, and reader-independent biomarkers, which are particularly valuable for indeterminate PI-RADS 3 lesions where subjective interpretation often leads to diagnostic uncertainty.³³ Therefore, the proposed histogram approach can serve as a complementary tool to PI-RADS v2.1, enhancing the reproducibility and accuracy of prostate MRI interpretation.

Interpretation of Individual Histogram Features

A direct comparison with a prior 3.0T study³⁴ reveals a notable entropy paradox: ADC entropy was a significant discriminator at 3.0T but non-significant at 5.0T (adjusted P = 0.952). This may be explained by three interrelated field-strength-specific mechanisms. First, ultra-high SNR at 5.0T reduces noise-driven entropy inflation. Second, improved spatial resolution enables more precise ROI segmentation, minimizing partial-volume effects.³⁵ Third, enhanced T2 and magnetic susceptibility effects at 5.0T stabilize diffusion signal homogeneity, narrowing entropy distribution differences between PCa and BPH.³⁶ Conversely, D* entropy showed a trend toward higher values in PCa (raw P = 0.016), although this did not remain significant after Holm-Bonferroni correction (adjusted P = 0.208). This reflects perfusion-related heterogeneity that may be more pronounced at ultra-high field,³⁷ while skewness of the f parameter was non-significant, likely due to the high consistency of perfusion fraction measurements at 5.0T.³⁸ Collectively, these observations indicate that histogram feature significance is highly field-strength dependent, and the entropy paradox reflects improved imaging fidelity at ultra-high field rather than a contradictory finding.

Field-strength-specific Effects of 5.0T on Quantitative Histogram Parameters

5.0T ultra-high-field MRI exerts unique, field-dependent effects on histogram metrics that are not observed at 3.0T, representing a key novel mechanistic contribution of this study. First, 5.0T MRI amplifies T2 relaxation time differences between BPH and malignant prostate tissue. Increased static field strength widens the relative gap in T2 values between benign and malignant tissue, enhancing between-group differences in T2 mapping histogram metrics (mean, median, 10th percentile⁾.³⁹ Second, 5.0T MRI improves diffusion signal fidelity. Reduced thermal noise and improved B₁⁺ homogeneity produce more stable ADC and D measurements, thereby reducing within-group histogram variance and enhancing measurement reproducibility.⁴⁰ Third, 5.0T MRI increases histogram binning precision. Higher spatial resolution minimizes partial-volume effects and increases voxel homogeneity, making histogram features (especially kurtosis, skewness, entropy) more biologically representative of true tissue microstructure rather than imaging nois.^41,42 These field-specific effects collectively suggest that the high diagnostic accuracy of our 5.0T-based model may be partly attributable to improved SNR, spatial resolution, and B₁⁺ homogeneity at ultra-high-field strength.

Study Limitations

This study has several limitations that should be acknowledged. First, this was a single-center, retrospective observational study with a relatively small sample size, which may limit the generalizability of our findings. Although 5-fold stratified cross-validation yielded a minimal optimism value of 0.008, indicating negligible overfitting in this cohort, the absence of an external validation cohort means the reported AUC values may represent optimistic estimates. Second, no direct intra-individual comparison between 5.0T and 3.0T MRI was performed in this study, so the independent contribution of field strength to the observed diagnostic performance cannot be definitively quantified. Third, over 70 statistical comparisons were performed, and although the Holm-Bonferroni correction for multiple comparisons was applied to control the family-wise error rate, residual type-I error cannot be fully excluded. Fourth, the study enrollment period was limited to six months (January–July 2024), which may introduce selection bias. Fifth, lesion diameter was restricted to ≥5 mm based on voxel count requirements for reliable histogram analysis, potentially excluding small clinically significant tumors. Sixth, manual ROI placement was inherently subjective, and no automated segmentation algorithm was employed in this study, which may affect the reproducibility of the quantitative histogram metrics.

Conclusion

In conclusion, this study demonstrates that ultra-high-field 5.0T multiparametric MRI combined with quantitative histogram analysis achieves high diagnostic accuracy for distinguishing PCa from BPH. This objective, reader-independent approach offers a promising complement to PI-RADS v2.1, with performance comparable to the clinical standard, particularly for reducing diagnostic uncertainty in indeterminate (PI-RADS 3) lesions.

Supplemental Material

Supplemental Material - Quantitative Histogram Analysis of 5.0T Multiparametric MRI for Discrimination Between Prostate Cancer and Benign Hyperplasia

Supplemental Material for Quantitative Histogram Analysis of 5.0T Multiparametric MRI for Discrimination Between Prostate Cancer and Benign Hyperplasia by Chengfeng Zheng, Sen Xing,, Xinghua Liu, Shaoxin Xiang, Ying Xiong, Huan Ma, Yixin Emu, Wenbing Zeng in Technology in Cancer Research & Treatment

Footnotes

ORCID iD

Chengfeng Zheng

Ethical Considerations

The study obtained ethical approval from the Medical Ethics Committee of Chongqing University Three Gorges Hospital, (Approval Batch Number: KS-2024072)and individual consent for this retrospective analysis was waived.

Author Contributions

Chengfeng Zheng: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Methodology, Formal analysis, Data curation, Conceptualization.

Sen Xing: Writing – review & editing, Visualization, Investigation, Conceptualization.

Xinghua Liu: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Conceptualization.

Shaoxin Xiang: Writing – review & editing, Data curation.

Ying Xiong: Data curation, Conceptualization.

Huan Ma: Visualization, Project administration.

Yixin Emu: Writing – review & editing, Visualization.

Wenbing Zeng: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Formal analysis.

All authors confirmed that they meet the ICMJE criteria for authorship. Each author has made substantial contributions to the conception, acquisition, analysis, or interpretation of the work; participated in drafting or critically revising the manuscript for important intellectual content; approved the final version to be published; and agrees to be accountable for all aspects of the work.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by the Chongqing City Key Medical Discipline Project (No. ZDXK202116).

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Supplemental Material

Supplemental material for this article is available online.

References

Siegel

Miller

Wagle

Jemal

. Cancer statistics, 2023. CA Cancer J Clin. 2023;73(1):17-48. doi:10.3322/caac.21763.

Collaborators

GBD

. The global, regional, and national burden of benign prostatic hyperplasia in 204 countries and territories from 2000 to 2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet Healthy Longev. 2022;3(11):e754-e776. doi:10.1016/S2666-7568(22)00213-6.

Schaeffer

Srinivas

Adra

, et al. Prostate Cancer, Version 4.2023, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw. 2023;21(10):1067-1096. doi:10.6004/jnccn.2023.0050.

Kim

Kwon

Joung

. Impact of benign prostatic hyperplasia and/or prostatitis onthe risk of prostate cancer in Korean patients. World J Mens Health. 2020;39(2):358-365. doi:10.5534/wjmh.190135.

Siegel

Miller

Jemal

. Cancer statistics, 2017. CA Cancer J Clin. 2017;67(1):7-30. doi:10.3322/caac.21387.

Moradi

Srinivasan

Clements

Batra

. Beyond the biomarker role: prostate-specific antigen (PSA) in the prostate cancer microenvironment. Cancer Metastasis Rev. 2019;38(3):333-346. doi:10.1007/s10555-019-09815-3.

Merriel

SWD

Pocock

Gilbert

, et al. Systematic review and meta-analysis of the diagnostic accuracy of prostate-specific antigen (PSA) for the detection of prostate cancer in symptomatic patients. BMC Med. 2022;20(1):54. doi:10.1186/s12916-021-02230-y.

Getaneh

Heijnsdijk

EAM

de Koning

. The comparative effectiveness of mpMRI and MRI-guided biopsy vs regular biopsy in a population-based PSA testing: a modeling study. Sci Rep. 2021;11(1):1801. doi:10.1038/s41598-021-81459-2.

Launer

Ellis

Scarpato

. A contemporary review: mpMRI in prostate cancer screening and diagnosis. Urol Oncol. 2024;42(11):337-345. doi:10.1016/j.urolonc.2024.05.012.

10.

Turkbey

Rosenkrantz

Haider

, et al. Prostate Imaging Reporting and Data System version 2.1: 2019 update of Prostate Imaging Reporting and Data System version 2. Eur Urol. 2019;76:340-351. doi:10.1016/j.eururo.2019.02.033.

11.

Chen

Meng

Cao

, et al. Histogram analysis of MR quantitative parameters: are they correlated with prognostic factors in prostate cancer. Abdom Radiol. 2024;49:1534-1544. doi:10.1007/s00261-024-04227-6.

12.

Muller

Shih

Sankineni

, et al. Prostate cancer: interobserver agreement and accuracy with the revised Prostate Imaging Reporting and Data System at multiparametric MRimaging. Radiology. 2015;277:741-750. doi:10.1148/radiol.2015142818.

13.

Le Bihan

Breton

Lallemand

Grenier

Cabanis

Laval-Jeantet

. MR imagingof intravoxel incoherent motions: application to diffusion and perfusion in neurologic disorders. Radiology. 1986;161:401-407. doi:10.1148/radiology.161.2.3763909.

14.

Meng

Gan

Chen

Wang

Liu

. Intravoxel incoherent motion predicts positivesurgical margins and Gleason score upgrading after radical prostatectomy for prostate cancer. Radiol Med. 2023;128:668-678. doi:10.1007/s11547-023-01645-2.

15.

Shi

Zhao

Zhu

, et al. Time-of-Flight Intracranial MRA at 3 T versus 5 T versus 7T: Visualization of distal small cerebral arteries. Radiology. 2022;306:207-217. doi:10.1148/radiol.220114.

16.

Pohmann

Speck

Scheffler

. Signal-to-noise ratio and MR tissue parameters in human brain imaging at 3, 7, and 9.4 tesla using current receive coil arrays. Magn Reson Med. 2015;75:801-809. doi:10.1002/mrm.25677.

17.

Zhang

Sheng

Yang

Dai

Zeng

. Higher field reduced FOV diffusion-weighted imaging for abdominal imaging at 5.0 Tesla: image quality evaluation compared with 3.0 Tesla. Insights Imaging. 2023;14:171. doi:10.1186/s13244-023-01513-7.

18.

Jiang

Sun

, et al. Stability and repeatability of diffusion-weighted imaging (DWI) of normal pancreas on 5.0 Tesla magnetic resonance imaging (MRI). Sci Rep. 2023;13:11954. doi:10.1038/s41598-023-38360-x.

19.

Zheng

Yang

Sheng

Dai

Zeng

. Renal imaging at 5 T versus 3 T: a comparison study. Insights Imaging. 2022;13:155. doi:10.1186/s13244-022-01290-9.

20.

Wolf

Wender

Etzioni

, et al. American Cancer Society guideline for the earlydetection of prostate cancer: update 2010. CA Cancer J Clin. 2010;60:70-98. doi:10.3322/caac.20066.

21.

Turkbey

Mani

Aras

, et al. Correlation of magnetic resonance imaging tumor volume with histopathology. J Urol. 2012;188:1157-1163. doi:10.1016/j.juro.2012.06.011.

22.

Carré

Raimondi

Botta

, et al. Effects of MRI image normalization techniques in prostate cancer radiomics. Phys Med. 2020;71:7-13. doi:10.1016/j.ejmp.2020.02.007.

23.

Ten Hove

Jorgensen

van der Ark

. Updated guidelines on selecting an intraclass correlation coefficient for interrater reliability, with applications to incomplete observational designs. Psychol Methods. 2022;29:967-979. doi:10.1037/met0000516.

24.

Bossuyt

Reitsma

Bruns

, et al. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. 2015;351:h5527. doi:10.1136/bmj.h5527.

25.

Zhang

Xue

. Risk stratification of prostate cancer using the combination of histogram analysis of apparent diffusion coefficient across tumor diffusion volume and clinical information: a pilot study. J Magn Reson Imaging. 2019;49(2):556-564. doi:10.1002/jmri.26235.

26.

Zheng

Yang

Liang

Rao

Dai

Zeng

. T2-weighted MRI and reduced-FOV diffusion-weighted imaging of the human pancreas at 5 T: a comparison study with 3 T. Med Phys. 2022;50:344-353. doi:10.1002/mp.15970.

27.

Yan

Zhu

, et al. The value of apparent diffusion coefficient values in predicting Gleason grading of low to intermediate-risk prostate cancer. Insights Imaging. 2024;15(1):137. doi:10.1186/s13244-024-01684-x.

28.

Dong

Wang

Shen

, et al. A combination of prostate volume and apparent diffusion coefficient can stratify patients with a PI-RADS score of 3 to reduce unnecessary prostate biopsies. Prostate. 2024;84:780-787. doi:10.1002/pros.24695.

29.

Xiong

Shen

Fan

, et al. 5 T versus 3 T MRI for prostate cancer: an intra-individual prospective comparison of image quality and diagnostic performance. Prostate Cancer Prostatic Dis. 2026. doi:10.1038/s41391-026-01073-z5. Epub ahead of print.

30.

Shen

Cheng

Liu

, et al. Reduced FOV T2-weighted imaging, 2.5D T2-weighted imaging, and high b-value diffusion-weighted imaging of the prostate at 5 T: a comparative study with 3 T. Abdom Radiol. 2025;50:5324-5337. doi:10.1007/s00261-025-04954-4.

31.

Deng

Liu

Chen

, et al. Comparison of 5T and 3T Biparametric MRI for Clinically Suspected Prostate Cancer: Image Quality, Lesion Visualization, and Artifacts. J Magn Reson Imaging. 2025;0(0):1-12. doi:10.1002/jmri.70171.

32.

, et al. Intravoxel incoherent motion diffusion-weighted imaging used to detect prostate cancer and stratify tumor grade: a meta-analysis. Front Oncol. 2020;10:1623. doi:10.3389/fonc.2020.01623.

33.

Ren

Peng

Gong

. Accuracy, intra-and inter-radiologist variability of PI-RADS v2.1 scoring for clinically significant prostate cancer detection. Quant Imaging Med Surg. 2025;15(8):7080-7089. doi:10.21037/qims-2025-37.

34.

İbrahim

Aydın

Hızlı

. The use of ADC histogram analysis in the diagnosis and determination of aggressiveness of peripheral zone prostate cancer. Pol J Radiol. 2025;90:374-383. doi:10.5114/pjr/205459.

35.

Aribisala

Blamire

. Comparative study of standard space and real space analysis of quantitative MR brain data. J Magn Reson Imaging. 2011;33(6):1503-1509. doi:10.1002/jmri.22576.

36.

Syversen

Elschot

Sandsmark

Bertilsson

Bathen

Goa

. Exploring the diagnostic potential of adding T2 dependence in diffusion-weighted MR imaging of the prostate. PLoS One. 2021;16(5):e0252387. doi:10.1371/journal.pone.0252387.

37.

Chang

Lin

Wong

, et al. IVIM parameters on MRI could predict ISUP risk groups of prostate cancers on radical prostatectomy. Front Oncol. 2021;11:659014. doi:10.3389/fonc.2021.659014.

38.

Chirra

Leo

Yim

, et al. Multisite evaluation of radiomic feature reproducibility and discriminability for identifying peripheral zone prostate tumors on MRI. J Med Imaging. 2019;6(2):024502. doi:10.1117/1.JMI.6.2.024502.

39.

Luo

Zhu

, et al. Comparison of 5T and 3T biparametric MRI for clinically suspected prostate cancer: image quality, lesion visualization, and artifacts. J Magn Reson Imaging. 2026;63(3):686-697. doi:10.1002/jmri.70171.

40.

Wei

Chen

Han

, et al. 5T magnetic resonance imaging: radio frequency hardware and initial brain imaging. Quant Imaging Med Surg. 2023;13(5):3222-3240. doi:10.21037/qims-22-945.

41.

JH Klinger, D Leithner, S Woo, M Weber, IV Vargas, ME Mayerhoefer . Reproducibility of radiomic features of the brain on ultrahigh-field MRI at 7 Tesla: a comparison of different segmentation techniques. Clin Radiol. 2026;92:107151. doi:10.1016/j.crad.2025.107151.

42.

Duron

Balvay

Vande

, et al. Gray-level discretization impacts reproducible MRI radiomics texture features. PLoS One. 2019;14(3):e0213459. doi:10.1371/journal.pone.0213459.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.34 MB