A nomogram using transition zone PSA density for detecting clinically significant prostate cancer in PI-RADS 3 lesions

Abstract

This study aimed to explore several transitional zonal (TZ)-adjusted parameters for the prediction of clinically significant prostate cancer (csPCa) that categorized as Prostate Imaging Reporting and Data System (PI-RADS) score 3 lesions. A total of 288 patients who underwent multiparametric MRI (mpMRI) and MRI/US targeted fusion biopsy between January 2017 and March 2024 were included in this retrospective study, which was randomly assigned as training dataset and validation dataset. Diagnostic performance of different variables was analyzed using receiver operating characteristics (ROC) and compared with area under ROC (AUC). Univariable and multivariable logistic regression analysis were used to determine significant variables, and a nomogram integrating these parameters was developed. The calculated AUC for prostate-specific antigen density (PSAD), TZ volume (TZV), PSAD based on TZ (TZ-PSAD), and TZ volume ratio (TZ-ratio) were 0.644 (95% CI 0.558–0.731), 0.710 (95% CI 0.610–0.810), 0.724 (95% CI 0.638–0.809), and 0.735 (95% CI 0.650–0.820), respectively. The Delong test demonstrated that TZ-PSAD and TZ-ratio outperformed PSAD, with P = 0.002 and P = 0.04. AUC for model integrating age, TZ-PSAD, and TZ-ratio was 0.834 (95% CI 0.764–0.904), which outperformed TZ-PSAD (P = 0.003) and TZ-ratio (P = 0.007) alone. Zonal-adjusted variables showed promising diagnostic accuracy for the detection of csPCa among lesions scored as PI-RADS category 3, and both TZ-PSAD and TZ-ratio outperformed the whole gland PSAD. Moreover, the model incorporating age, TZ-PSAD, and TZ-ratio could further improve the overall diagnostic performance.

Introduction

Prostate cancer is the second most prevalent malignancy affecting men worldwide, necessitating precise diagnostic tools to ensure timely and accurate detection^1,2. Although mpMRI-based PI-RADS has exhibited high diagnostic performance and widespread adoption for PCa risk stratification, which showed high sensitivity but lower specificity, resulting in many unnecessary biopsies and overdiagnosis^3–5. Particularly, PI-RADS 3 lesions, which are described as “equivocal” or “intermediate” in the presence of csPCa, often referred to biopsy even though follow-up strategy could be an acceptable option in low-risk patients. Currently, no standardized management strategies for surveillance or repeat biopsy for these intermediate or undetermined lesions. A previous meta-analysis reported that the proportion of csPCa in PI-RADS score 3 lesions varied widely among studies, ranging from 3.4 to 46.5%⁶. This variability underscores the difficulty in managing these cases and highlights the need for more reliable diagnostic markers.

In recent years, PSAD has been extensively studied as both an independent predictor and in combination with other clinical variables for PCa^7–9. However, when PSAD is used alone, its predictive performance remains suboptimal. For instance, in PI-RADS 3 lesions, the pooled sensitivity and specificity of PSAD at a threshold of 0.15 ng/mL/mL were reported to be only 0.61 and 0.69, respectively¹⁰. To improve diagnostic accuracy, Kalish et al. first proposed transition zone (TZ)-based PSAD in 1994, recognizing that benign prostatic hyperplasia (BPH) predominantly originates in the TZ, whereas the peripheral zone (PZ) and central zone contribute minimally¹¹. Subsequent studies have suggested that TZ-PSAD is a stronger predictor of PCa than whole-gland PSAD. However, earlier studies relied on transrectal ultrasound (TRUS) for prostate volume estimation, which has lower soft-tissue contrast than MRI, potentially limiting the accuracy of volumetric measurements^12–14. Recently, studies have shown that zonal-adjusted variables such as TZV, peripheral zone volume (PZV), TZ-ratio, and PZ-ratio demonstrated promising ability for detecting csPCa, and superior to whole gland PSAD^12,15,16. However, these studies were primarily focused on patients with a PSA level of 4–20 ng/ml. Therefore, this study intended to look into the diagnostic performance of using TZ-PSAD and TZ-ratio for differentiating csPCa in PI-RADS score 3 lesions.

Materials and methods

Patient selection

This study was conducted in compliance with the guiding principles of the Declaration of Helsinki and was approved by the institutional review board (IRB) of The Affiliated Suzhou Hospital of Nanjing University Medical School, with requirement of written informed consent was waived by the IRB of our institution (The Affiliated Suzhou Hospital of Nanjing University Medical School) because of the retrospective nature. We retrospectively reviewed TZ lesions stratified as PI-RADS 3 from the electric medical records from our institution between January 1, 2017 and March 31, 2024, only those that underwent mpMRI examination and subsequently MRI/US fusion targeted biopsy were included. Two radiologists assessed these lesions independently according to PI-RADS v2.1 again, who were blinded to pathological results. Exclusion criteria as follows: (1) history biopsy or treatment as PCa; (2) only underwent systematic biopsy; and (3) severe artifacts on DWI images. Eventually, 288 lesions (one lesion per patient) were included in the study population, which was divided into the training cohort (202 lesions) and the validation cohort (86 lesions) with a ratio of 7:3. The patient selection flowchart is presented in Fig. 1.

Fig. 1.

Flowchart of the study population with the exclusion criteria. csPCa, clinically significant prostate cancer.

MRI acquisition and interpretation

Prostate MRI was performed on a 3T system (MAGNETOM Skyra, Germany), with a 32-channel phased array coil. The MRI acquisition protocols were optimized according to the PI-RADS recommendations, which included the following sequences: axial turbo spin echo (TSE) T2 weighted image (repetition time [TR] = 4400 ms; echo time [TE] = 96 ms; slice thickness = 3.0 mm; scanning field of view (FOV) = 230 mm×230 mm; matrix = 320 × 224), sagittal TSE T2WI (TR = 3800 ms; TE = 117 ms; slice thickness = 3.0 mm, FOV = 230 mm×230 mm, matrix = 320 × 256), axial DWI (TR = 6570 ms, TE = 88 ms, layer thickness = 3.0 mm, FOV = 320 mm×224 mm, matrix = 192 × 192) and multiple b values (b = 0, 100, 1000, 2000 s/mm²), and DCE (TR = 6570 ms, TE = 1770 ms, layer thickness = 3.0 mm, FOV = 260 mm×260 mm, matrix = 192 × 154).

Prostate biopsy procedure

All patients received systematic TRUS biopsy at least four weeks following their MRI, with a double sextant approach that involved obtaining 12 cores. This sampling targeted both lateral and medial regions at the apex, mid-gland, and base of each prostate half. For individuals with suspicious findings on mpMRI, targeted fusion biopsies were performed using an ESAOTE Mylab Twice color Doppler ultrasound system, equipped with a 7.5-MHz transrectal end-fire probe and Real-time Virtual Sonography fusion technology. Identified lesions were annotated on the MRI T2WI by the radiologists, and an arrow was placed pointing to the approximate center of the target, assigning the lesion to the PZ or TZ based on PI-RADS v2.1 criteria. Targeted biopsy was performed in a transperineal approach while taking at least two cores (axial and sagittal planes) for individual lesion. These procedures were conducted by a urologist with 16 years of experience in prostate biopsies. All tissue samples were evaluated by a genitourinary pathologist with over 14 years of experience, who was blinded to MRI data. Gleason scores (GS) were assigned in accordance with the 2014 consensus guidelines of the International Society of Urological Pathology (ISUP): ISUP 1 = GS 3 + 3; ISUP 2 = GS 3 + 4; ISUP 3 = GS 4 + 3; ISUP 4 = GS 4 + 4; ISUP 5 = GS 9–10. Biopsy outcomes were analyzed to identify the location and GS of the index tumor that defined as the lesion with the highest GS. In the case of multiple cores presented the highest GS, the core with the highest tumor-specific percentage was identified as the location of an index tumor. Clinically significant prostate cancer (csPCa) was defined as GS ≥ 7 (ISUP ≥ 2)^17–19. Prostate whole-gland volume and TZ volume were measured based on T2-weighted MRI and calculated using the ellipsoid volume formula: transverse width × transverse length × longitudinal height × π/6. In this study, all TZ lesions scored as PI-RADS 3 were included, regardless of whether they were identified via MRI/US fusion targeted biopsy or TRUS biopsy, and irrespective of the presence or absence of positive systematic biopsies from the PZ.

Statistical analysis

Univariable logistic regression analysis was performed for each potential clinical parameter to inspect significant predictors of csPCa, including age, PSA level, whole prostate volume and transitional volume, and TZ-ratio (defined as TZ volume divided by whole gland volume). Subsequently, multivariable logistic regression analysis was conducted to explore the significant clinical predictors, which were used to construct the final prediction nomogram. The AUC was used to determine the overall diagnostic performance, and the best was defined as the largest one. Comparisons between AUCs were performed using Delong’s test, and the sensitivity and specificity were decided with the Youden index. Additionally, the decision curve analysis (DCA) was performed to evaluate the clinical usefulness and benefits of the integrated nomogram at different threshold probabilities, and a calibration curve was plotted to explore the predictive accuracy of the combined model. All analysis was performed with R 4.3.2, with a P < 0.05 indicated statistically significant.

Results

Patient characteristics

The detailed patient characteristics are demonstrated in Table 1. We investigated 288 TZ lesions scored as PI-RADS 3 with version 2.1, MRI/US fusion targeted biopsy and systematic TRUS biopsy revealed that 56 lesions (19.44%) were csPCa. In these malignancies, 41 were assigned to the training group and 15 were assigned to the validation group. No significant difference between the training set and validation set regarding age, PSA, PV, PSAD, TZV, TZ-PSAD, and TZ-ratio, with a P value of 0.27–0.93. However, significant differences in these variables were observed between csPCa and non-csPCa in both the training group and validation group (P < 0.05).

Table 1.

Characteristics of Patients.

Variable	Training (n = 202)			Validation (n = 86)
	csPCa (n = 41)	Non-csPCa (n = 161)	P	csPCa (n = 15)	Non-csPCa (n = 71)	P
Age (Years, mean ± SD)	73.9/6.26	69.66/8.32	0.003	74.4/6.22	68.85/7.46	0.009
PSA (ng/mL, median [IQR])	10.6/7.46–18.37	10.8/7.92–16.16	0.06	12/8.73-32	10.38/8.24–16.64	0.3
PV (ml, median [IQR])	73.83/49.26–113.2	91.81/70.91–123.6	0.15	71.32/51.09-117.66	90.44/73.94-128.46	0.12
PSAD (ng/mL/mL, median [IQR])	0.15/0.12–0.25	0.12/0.08–0.17	0.002	0.17/0.12–0.33	0.12/0.09–0.17	0.014
TZV (ml, median [IQR])	46.59/24.46–82.62	69.31/49.61-102.29	< 0.001	41.4/29.14–97.21	67.12/48.59-105.34	0.03
TZ-PSAD (ng/ml/ml, median [IQR])	0.34/0.21–0.45	0.16/0.11–0.26	< 0.001	0.31/0.21–0.57	0.17/0.13–0.25	0.004
TZ-ratio (mean ± SD)	0.58/0.39–0.66	0.73/0.66–0.87	< 0.001	0.60/0.46–0.79	0.72/0.65–0.88	0.008

csPCa, clinically significant prostate cancer; IQR, interquartile range; PSA, prostate-specific antigen; PSAD, prostate-specific antigen density; PV, prostate volume; SD, standard deviation; TZ-PSAD, transitional zone prostate-specific antigen density; TZ-ratio, transitional zone volume ratio; TZV, transitional zone volume.

ROC analysis for PSAD, TZ-PSAD, and TZ-ratio

The calculated AUC for whole gland PSAD, TZV, TZ-PSAD, and TZ-ratio were 0.644 (95% CI 0.558–0.731), 0.710 (95% CI 0.610–0.810), 0.724 (95% CI 0.638–0.809), and 0.735 (95% CI 0.650–0.820) in the training group, respectively (Fig. 2). Delong’s test revealed that there was no statistically different between TZ-PSAD, TZV, and TZ-ratio regarding overall diagnostic accuracy. However, ROC analysis demonstrated that TZ-PSAD and TZ-ratio had higher diagnostic performance than PSAD (P = 0.002 and P = 0.04, respectively). The optimal threshold of PSAD for detecting csPCa among PI-RADS 3 lesions was 0.11 ng/ml/ml, where the sensitivity and specificity were 48.8% (95% CI 32.9%-64.9%) and 85.1% (95% CI 78.6%-90.2%), respectively. Regarding TZ-PSAD, at a threshold of 0.21 ng/ml/ml, the sensitivity was 65.9% (95% CI 49.4%-79.9%), with specificity of 75.8% (95% CI 68.4%-82.2%). The optimal threshold for TZ-ratio was 0.659, where the sensitivity was 65.9% (95% CI 49.4%-79.9%), with specificity of 72.7% (95% CI 65.1%-79.4%).

Fig. 2.

ROC analysis for prediction of clinically significant prostate cancer. (A) training group; (B) validation group. PSAD, prostate-specific antigen density. TZ-PSAD, transitional zone prostate-specific antigen density; TZ-ratio, transitional zone volume ratio.

Univariate and multivariate regression analyses for CsPCa

The univariate logistic regression analysis demonstrated that age, PSAD, TZ-PSAD, PV, TZV, and TZ-ratio were independent clinical variables for differentiating csPCa (Table 2). However, in the subsequent multivariable logistic regression analysis PSAD, PV, and TZV were excluded. Therefore, only age, TZ-PSAD, and TZ-ratio were retained to construct the model, which generated an AUC of 0.834 (95% CI 0.764–0.904). Delong’s test showed that the combination model resulted in significant improvement as compared to either TZ-PSAD (P = 0.003) and TZ-ratio (P = 0.007). The decision curve analysis on the validation data set showed an added net benefit of using the integrated model over PSAD, TZ-PSAD, and TZ-ratio (Fig. 3). The calibration curves also showed good concordance between the predicted probabilities of the model for csPCa and the observed outcomes, which are demonstrated in Fig. 4. In the validation set, the calculated AUC of for integrated model showed a favorite outcome, with sensitivity and specificity of 84.5% (95% CI 59.5%-98.3%) and 66.7% (95% CI 54.0%-77.0%).

Table 2.

Univariate and multivariate logistic regression Analysis.

Variable	OR	95% CI	SE	P
Univariate Logistic Regression Analysis
Age	1.07	1.02–1.12	0.03	0.003
PSAD	21.43	2.28-201.12	24.48	0.007
TZ-PSAD	31.39	5.75-171.37	27.18	< 0.001
PV	0.99	0.98-1.00	0.004	0.046
TZV	0.98	0.97–0.99	0.01	0.001
TZ-ratio	0.001	0.0001–0.015	0.002	< 0.001
Multivariate Logistic Regression Analysis
Age	1.11	1.04–1.17	0.03	0.001
TZ-PSAD	7.28	0.98–54.34	7.47	0.04
TZ-ratio	0.002	0.0001-0.04	0.003	< 0.001

AUC, area under the ROC curve; CI, confidence interval; OR, odd ratio; PSAD, prostate-specific antigen density; SE, standard error; TZ-PSAD, transition zone prostate-specific antigen density; TZ-ratio, transition zone volume ratio; TZV, transition zone prostate volume.

Fig. 3.

Decision curves analyses for prediction of clinically significant prostate cancer.

Fig. 4.

Calibration curves.

Discussion

In this study, our findings demonstrated that TZ-PSAD and TZ-ratio had promising ability for the detection of csPCa in TZ lesions scored as PI-RADS 3. Moreover, both two zonal-adjusted variables significantly superior to the whole gland PSAD. Recent studies have demonstrated that PSAD is a promising predictor for PCa, and a cutoff value of 0.15 ng/ml/ml was recommended as the threshold. Nevertheless, a meta-analysis revealed that for PI-RADS score 3 lesions, PSAD merely showed moderate accuracy. In the current study, the best cutoff value of PSAD for detecting csPCa was 0.11 ng/ml/ml, where the calculated AUC was 0.644, with sensitivity and specificity of 66.5% and 75.6%. By comparison, two zonal-adjusted variables of TZ-PSAD and TZ-ratio yielded AUC of 0.724 and 0.735, both outperformed whole gland PSAD (P = 0.002 and P = 0.04). In a previous study, Schneider et al. performed a comparison between whole-gland PSAD and TZ-PSAD for the detection of PCa, they observed that the latter showed a stronger correlation with PCa²⁰. In their study, although the difference in AUC between these two parameters not reached statistically significant (AUC 0.72 vs. 0.69, p = 0.059), it should be noted that this comparison was performed between lesions of Gleason score 3 + 4 and 4 + 3. Several other studies also assessed the zonal-adjusted clinical variables for the detection of csPCa, which all showed favorable results as compared to PSAD^12–14,16. However, in some earlier studies, prostate volume was measured with transrectal US imaging, which has inferior soft tissue contrast to MRI, potentially affecting measurement accuracy and subsequent predictive performance. Despite this, findings from these studies remain showed the potential capacity of TZ-PSAD and TZV for predicting PCa^13,14,16,21. In a more recent study, Hamm et al. showed that TZ-PSAD significantly outperformed whole gland PSAD for GG ≥ 2 PCa, with AUC 0.66 vs. 0.62, P = 0.023²². To our knowledge, our study is the first to focus on TZ-adjusted clinical variables for PI-RADS 3 lesions. According to our analysis, the best threshold of TZ-PSAD for differentiating csPCa was 0.21 ng/ml/ml, which was consistent with previous studies (0.22 ng/ml/ml)²⁰. By combining 3 clinical parameters derived from multivariable logistic analysis, substantial improvement was observed as compared to using TZ-PSAD or TZ-ratio alone, and which was tested in an independent validation cohort.

PI-RADS score 3 lesions indicate an indeterminate probability of csPCa, posing a clinical challenge in balancing the need to avoid unnecessary biopsies while ensuring accurate PCa detection. For these equivocal lesions, PSAD has demonstrated only moderate accuracy, and its optimal cutoff value varies widely across studies, limiting its reliability as a independent predictor. Some studies explored the application of radiomics for these indeterminate lesions; however, in most of these studies the results were tested with the internal set, and the absence of external validation prevented the generalization of this method. At present, the management of PI-RADS 3 lesions remains challenging because of the broad spectrum of associated csPCa and the lack of a unified consensus on their evaluation. Compared to the complex methodologies of radiomics, our study developed a predictive model that achieved promising diagnostic performance using a more straightforward and clinically feasible approach.

The hypothesis behind using TZ-PSAD and the TZ-ratio as predictors for PCa is based on the fact that benign prostatic hyperplasia primarily arises in the transition zone. Therefore, PSA variations due to benign prostatic hyperplasia are more likely to originate from the hypertrophied glands in this zone. Consequently, other prostate zones should remain relatively constant and less affected by benign prostatic hyperplasia. Adjusting PSA levels to exclude variations from other prostate zones may improve of ability of PSA for detecting csPCa. Since Kalish et al. proposed using TZ-PSAD as an independent clinical variable for the detection of PCa in the transition zone, several studies have highlighted the potential diagnostic performance of zonal-associated variables¹¹. In addition to TZ, research has indicated that peripheral zone-adjusted PSAD also demonstrated potentiality in the prediction of csPCa in PZ lesions^15,23. Chen et al. demonstrated that csPCa located in PZ, the AUC for PZ-adjusted PSAD was significantly better than PSAD (AUC 0.824 vs. 0.792, P < 0.01) and the PZ-ratio (0.717, P < 0.01). In another study, Huang et al. revealed that PZ-adjusted PSAD outperformed the whole-gland PSAD (AUC 0.812 vs. 0.788, P < 0.01)²³. Nonetheless, the effectiveness of zone-adjusted PSAD remains debated. Elliott et al. reported no significant advantage of transition zone density over prostate-specific antigen density in predicting any PCa (P = 0.056); however, for csPCa, TZ-PSAD demonstrated superior performance compared to PSAD, with an AUC of 0.727 vs. 0.673 (P < 0.001)²⁴.

Our study has several limitations. First, this is a single-center, retrospective study, which may affect the generalization of our conclusion. Second, the study sample was relatively small. However, as our study was focused on PI-RADS score 3 lesions in TZ, which only account for a limited proportion of all patients. Third, the classification of lesions according to PI-RADS was subjectively by radiologists, thus the results primarily depended on the radiologist’s personal experience. In the current study, the experience of the readers is limited to 2 and 6 years, respectively. Radiologists with different experiences may score some cases in another manner. Last, in our study csPCa was defined as a Gleason score of 3 + 4 or higher; however, it is important to note that there is no universally accepted definition of csPCa. Different studies have adopted varying criteria, including number of positive biopsies is 20% or more of the total, maximum cancer length of biopsy core is 5 mm or more. This variability may limit the comparability of our findings with other studies and affect the generalizability of our results across different clinical contexts.

Conclusions

Our study revealed that several prostate volume-related variables, including TZV, TZ-PSAD, and TZ-ratio, were independent predictors of csPCa in PI-RADS score 3 lesions. Moreover, TZ-PSAD and TZ-ratio outperformed whole-gland PSAD for the detection of csPCa. Notably, integrating TZ-PSAD, TZ-ratio, and age significantly improved predictive accuracy, highlighting the potential of these parameters in refining risk stratification for patients with equivocal MRI findings.

Author contributions

Wang Cheng and Qiu Yi had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Zheng Borui and Han Jing.Acquisition, analysis, or interpretation of data: All authors. Drafting of the manuscript: Zhu Weiwei. Statistical analysis: Han Jing and Zhu Weiwei. Administrative, technical, or material support: Zheng Borui. Supervision: Qiu Yi.

Funding

None.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Written informed consent was waived by the Institutional Review Board.

Consent for publication

Patients’ data were anonymized. Not applicable.

Competing interests

The authors declare no competing interests.