Skip to main content

Automated pelvic MRI measurements associated with urinary incontinence for prostate cancer patients undergoing radical prostatectomy



Pelvic morphological parameters on magnetic resonance imaging (MRI), such as the membranous urethral length (MUL), can predict urinary incontinence after radical prostatectomy but are prone to interobserver disagreement. Our objective was to improve interobserver agreement among radiologists in measuring pelvic parameters using deep learning (DL)-based segmentation of pelvic structures on MRI scans.


Preoperative MRI was collected from 167 prostate cancer patients undergoing radical prostatectomy within our regional multicentric cohort. Two DL networks (nnU-Net) were trained on coronal and sagittal scans and evaluated on a test cohort using an 80/20% train-test split. Pelvic parameters were manually measured by three abdominal radiologists on raw MRI images and with the use of DL-generated segmentations. Automated measurements were also performed for the pelvic parameters. Interobserver agreement was evaluated using the intraclass correlation coefficient (ICC) and the Bland–Altman plot.


The DL models achieved median Dice similarity coefficient (DSC) values of 0.85–0.97 for coronal structures and 0.87–0.98 for sagittal structures. When radiologists used DL-generated segmentations of pelvic structures, the interobserver agreement for sagittal MUL improved from 0.64 (95% confidence interval 0.28–0.83) to 0.91 (95% CI 0.84–0.95). Furthermore, there was an increase in ICC values for the obturator internus muscle from 0.74 (95% CI 0.42–0.87) to 0.86 (95% CI 0.75–0.92) and for the levator ani muscle from 0.40 (95% CI 0.05–0.66) to 0.61 (95% CI 0.31–0.78).


DL-based automated segmentation of pelvic structures improved interobserver agreement in measuring pelvic parameters on preoperative MRI scans.

Relevance statement

The implementation of deep learning segmentations allows for more consistent measurements of pelvic parameters by radiologists. Standardized measurements are crucial for incorporating these parameters into urinary continence prediction models.

Key points

• DL-generated segmentations improve interobserver agreement for pelvic measurements among radiologists.

• Membranous urethral length measurement improved from substantial to almost perfect agreement.

• Artificial intelligence enhances objective pelvic parameter assessment for continence prediction models.

Graphical Abstract


Urinary incontinence is a potential complication in patients with prostate cancer (PCa) undergoing robot-assisted radical prostatectomy (RARP). The 12-month postoperative prevalence of urinary leakage and its interference with daily life were reported by 36% and 17% of men, respectively [1]. Several patient-related variables such as age, body mass index, baseline urinary function, surgical factors, and magnetic resonance imaging (MRI)-based measurements have been found to be associated with post-prostatectomy continence recovery [2,3,4]. MRI-based measurements include parameters related to the prostate, the urethra, and the pelvic musculoskeletal system, such as the membranous urethral length (MUL), the levator ani muscle (LAM) thickness, and the intravesical prostatic protrusion length (IPPL). Among these measurements, the MUL has demonstrated the highest predictive value, with greater preoperative MUL serving as an independent prognostic factor for continence recovery [2, 5].

The measurement of the MUL is typically performed manually on T2-weighted prostate MRI. It involves determining the distance of the membranous urethra from the inferior border of the prostate apex to the superior border of the penile bulb in either the coronal or sagittal plane. However, manual assessment is prone to considerable interobserver disagreement, with varying intra-class correlation coefficient (ICC) values from 0.37 to 0.57 [6]. To improve interobserver agreement, Veerman et al. [7] investigated the use of standardized anatomical landmarks for MUL measurements, resulting in improved interobserver agreement with ICC values ranging from 0.63 to 0.84 among three radiologists within a single center.

Despite this improvement, Boellaard et al. [8] emphasized the need for artificial intelligence (AI) to further improve interobserver agreement through automated measurements of pelvic parameters. AI models can be used to segment relevant structures and improve the consistency and efficiency of radiologists on prostate MRI [9]. Small deviations in pelvic measurements can have a significant impact on the predictive probability of continence recovery when incorporating them into urinary incontinence prediction models and can influence the treatment choices of PCa patients [10].

Therefore, our study aims to standardize pelvic floor measurements using AI-driven segmentations, enabling automated measurements of the prostate, membranous urethra, and pelvic musculoskeletal structures. We assessed interobserver agreement among radiologists using both manual assessment on raw MRI images and AI-aided visualization of relevant pelvic structures. Additionally, we compared the pelvic measures obtained through the fully automated workflow with those acquired using the manual approach.


Data collection

MRI data for this study was retrospectively extracted from a prospective registry (NCT04228211) where patients had provided informed consent for data sharing. This prospective registry received approval from the institutional review and ethics board of the University Medical Center Utrecht (19–692/M). The study population included 167 PCa patients who underwent RARP within our regional multicentric prospective registry between March 2020 and November 2022. Exclusion criteria included patients with T4 tumors (n = 0). We collected preoperative 1.5-T or 3-T prostate MRI data from five Dutch hospitals within our region. Furthermore, MRI scans of three additional patients from three hospitals outside our region were also included in the study. We included the T2-weighted turbo-echo sequences in the coronal and sagittal directions. One sagittal MRI scan had to be excluded due to the presence of motion artifacts. The MRI sequence parameters for each regional hospital are listed in Table 1.

Table 1 MRI sequence parameters of coronal and sagittal T2-weighted imaging within our regional muticentric cohort

The following data were extracted from the Utrecht Prostate Cohort database: age at diagnosis, prostate-specific antigen (PSA) at diagnosis, biopsy grade group, Prostate Imaging Data and Reporting System (PI-RADS) version 2 score, clinical stage, and the pathological stage after RARP (Table 2).

Table 2 Patient characteristics

Pelvic structures

In the coronal plane, the delineated structures encompassed the prostate, corpus spongiosum, membranous urethra, left and right obturator internus muscle (OIM), and the left and right levator ani muscle (LAM). In the sagittal plane, the delineated pelvic structures included the prostate, bladder, corpus spongiosum, and the membranous urethra. The anatomical definition of the pelvic structures was first discussed in a group meeting with three abdominal radiologists (F.W., E.H., and C.T.N., with nine, nine, and seven years of experience, respectively) based on clinical relevance and current literature. The membranous urethra was defined to be delineated from the lower border of the peripheral zone of the prostate to the upper border of the corpus spongiosum. The lumen of the urethra was delineated in the midsagittal and midcoronal slices for each MRI scan.

All MRI scans were randomly divided into a training and test cohort at an 8:2 ratio after stratification by MRI location to ensure a balanced representation in both cohorts. A single annotator (I.B.) performed the delineation of pelvic structures and controlled 70 scans of the training cohort. These 70 scans were divided into three subsets, with each subset being controlled by one radiologist. The feedback received from these radiologists was then integrated into the remaining cases of both the training and test cohorts.

Model development

Two three-dimensional convolutional neural networks (CNNs) were employed for training with a fivefold cross-validation for the coronal and sagittal directions, enabling automated segmentation of pelvic structures. We selected the state-of-the-art network architecture nnU-Net due to its automated pipeline that encompasses preprocessing to postprocessing. nnU-Net is an open-source model and has been successfully applied on many MRI-based anatomical sites [11], including small elongated neurovascular structures on prostate MRI [12]. Both nnU-Net models were trained for 1,000 epochs with a batch size of 2, utilizing a patch size of 24 × 256 × 256 and pooling operations of 2 along the first axis and 6 along the second and third axes.

Pelvic measurements

Post-processing steps were performed to automate pelvic measurements on T2-weighted coronal and sagittal scans using a set of functions in Python 3.9, collecting the distances in millimeters (mm). The pelvic structures and measurements are shown in Fig. 1.

  1. 1.

    Coronal MUL: The largest segmentation in coronal plane was selected to determine the coronal MUL. We located the centroid of the segmentations in each axial slice and generated a line positioned at the center of the MUL for each axial slice to calculate the length in coronal direction. In cases where the membranous urethra was divided into two parts across adjacent slices, with one part connected to the prostate and the other part connected to the corpus spongiosum, both parts were considered for calculating the MUL.

  2. 2.

    Sagittal MUL: The largest segmentation in sagittal plane was selected to determine the sagittal MUL. The length was calculated with the same post-processing steps as for the coronal MUL.

  3. 3.

    IPPL: The length of the intravesical prostatic protrusion (IPPL) was determined by finding the intersection between the bladder and the prostate. This intersection was obtained by performing a one-pixel dilation of the bladder. A line was then formed connecting the outermost points of the intersection. The IPPL was computed as the maximum perpendicular distance between this line and the contour of the prostate.

  4. 4.

    LAM and OIM thickness: The maximum thickness of the left and right LAM and OIM was determined by compiling a list of the shortest distances between each point of the muscles’ centerline and its border in the coronal direction. The maximum distance from this list was selected and doubled to obtain the muscles' diameter.

Fig. 1
figure 1

Pelvic structures and pelvic measurements in coronal direction (a) and sagittal direction (b) on T2-weighted images. Pelvic structures include the bladder (purple), corpus spongiosum (green), prostate (yellow), membranous urethra (red), obturator internus muscle (dark blue), and levator ani muscle (cyan)

Model evaluation

The DL-generated segmentations of the pelvic structures were evaluated on the test cohort using volumetric Dice similarity coefficient (DSC), mean surface distances (MSD), and Hausdorff distances (HD95), utilizing the DeepMind Python package ( Volumetric DSC measures spatial overlap, MSD represent average distances, and HD95 capture maximum surface distances (95th percentile).

The pelvic MRI parameters (i.e., coronal MUL, sagittal MUL, IPPL, OIM thickness and LAM thickness) were evaluated on the test cohort in two sessions by three abdominal radiologists (F.W., E.H., C.T.N.). The first session comprised manual measurements on raw MRI images in coronal and sagittal direction and the second session involved manual measurements on MRI images with DL-generated segmentations of pelvic structures in ITK-SNAP (version 3.6.0; The evaluation of the second session was conducted at least seven days after the first evaluation, and the patient order was mixed to avoid recall bias. The radiologists were blinded to patients’ clinical data.

The absolute interobserver agreement for each measurement was determined by applying a two-way random effect analysis with the intraclass correlation coefficient (ICC). ICC values ≤ 0 indicate no agreement, 0.01–0.20 as none to slight agreement, 0.21–0.40 as fair agreement, 0.41–0.60 as moderate agreement, 0.61–0.80 as substantial agreement, and 0.81–1.00 as almost perfect agreement [13]. The interobserver agreement for each measurement was visualized in Bland–Altman plots between the manual and the AI-aided measurements (Fig. 2). Non-normally distributed data were presented as median with interquartile range (IQR). Wilcoxon-signed rank tests were conducted and p < 0.050 was considered statistically significant.


Segmentation performance

Table 3 presents the DL segmentation performances in the coronal and sagittal planes. The median DSC values ranged from 0.85 to 0.97 for coronal structures and from 0.87 to 0.98 for sagittal structures. The highest DSC value was observed for the bladder (DSC 0.98; IQR 0.97–0.98), while the lowest DSC value was found for the membranous urethra (DSC 0.85; IQR 0.76–0.92) in the coronal direction. The median MSD values ranged from 0.10 to 0.31 mm, with the lowest value observed for the membranous urethra in the sagittal direction (MSD 0.10 mm; IQR 0.05–0.14). Additionally, the DL network prediction time for the sagittal test cohort was 2 min and 40 s (approximately 5 s per scan) and for the coronal test cohort 3 min and 5 s (approximately 6 s per scan).

Table 3 Deep learning segmentation performances

Pelvic measurements

The pelvic measurements are presented in Table 4 for the three different measurement approaches: manual, AI-aided, and automated. The automated approach resulted in significantly higher values for the IPPL, LAM thickness, and OIM thickness (p < 0.001) and showed comparable values for the coronal MUL (p = 0.091) and sagittal MUL (p = 0.606) in comparison to the AI-aided approach. Sagittal MUL was highest for the manual approach (14.94 mm; IQR 12.08–17.73), followed by the automated approach (14.32 mm; IQR 12.50–17.12), and the AI-aided approach (14.04; IQR 11.63–16.97). Coronal MUL measurements were found to be lower, with a median value of 13.78 mm for the manual and AI-aided approach, and 14.06 for the automated approach. The median LAM thickness ranged from 8.56 to 11.14 mm, and the OIM thickness ranged from 17.92 to 19.77 mm, with the automated approach yielding the highest median values for both measurements.

Table 4 Comparison of pelvic measurements on T2-weighted MRI: manual, AI-aided and automated workflows with intraclass correlation coefficients

Interobserver agreement

The ICC values ranged from 0.40 to 0.88 for the manual approach and from 0.61 to 0.91 for the AI-aided approach (Table 4). The highest interobserver agreement was observed for sagittal MUL (ICC 0.91, 95% CI 0.84–0.95) with the AI-aided approach, indicating almost perfect agreement. The manual approach showed substantial agreement (ICC 0.64, 95% CI 0.28–0.83). The lowest interobserver agreement was found for LAM thickness, with the manual approach demonstrating moderate agreement (ICC 0.40, 95% CI 0.05–0.66) and the AI-aided approach showing substantial agreement (ICC 0.61, 95% CI 0.31–0.78). A Bland–Altman plot of sagittal MUL is presented in Fig. 2, showing the median difference between the manual and AI-aided approaches. The Bland–Altman plots of the other pelvic measurements are shown in Supplementary Figs. 14.

Fig. 2
figure 2

Bland–Altman plots of measured membranous urethral length (MUL) in the sagittal direction for manual assessment and artificial intelligence (AI)-aided assessment. The black dashed line represents the mean difference and the red dashed lines represent the upper and lower 95% limits of agreement


To our knowledge, this is the first study that developed and evaluated an automated workflow to measure pelvic floor parameters associated with post-prostatectomy incontinence. The implementation of DL-generated segmentations of pelvic structures resulted in higher interobserver agreement among radiologists. Both coronal and sagittal MUL measurements demonstrated improved interobserver agreement, with the coronal MUL increasing from an ICC value of 0.69 (95% CI 0.51–0.82) to 0.90 (95% CI 0.82–0.94) and the sagittal MUL improving from 0.64 (95% CI 0.28–0.83) to 0.91 (95% CI 0.84–0.95), indicating a shift from substantial to almost perfect agreement. Veerman et al. [7] reported a comparable ICC value of 0.63 (95% CI 0.28–0.81) for the sagittal MUL measurements on raw images but achieved an improved ICC value of 0.84 (95% CI 0.66–0.91) with the implementation of consistent anatomical definitions. In our study, DL-generated segmentations resulted in a higher ICC value of 0.91 (95% CI 0.84–0.95), demonstrating the potential of AI to further improve consistency in the measurements.

Additionally, the use of DL-generated segmentations improved interobserver agreement for the musculoskeletal structures. The OIM thickness had an ICC value of 0.74 (95% CI 0.42–0.87) for the manual assessment and 0.86 (95% CI 0.75–0.92) for the AI-aided assessment, indicating an improvement from substantial agreement to almost perfect agreement. The LAM thickness showed moderate agreement for the manual approach (ICC 0.40; 95% CI 0.05–0.66) and improved to substantial agreement for the AI-aided approach (ICC 0.61; 95% CI 0.31–0.78). However, IPPL measurements demonstrated no improvement with DL-generated segmentations, with ICC values of 0.88 (95% CI 0.81–0.94) and 0.86 (95% CI 0.77–0.93) for the manual and AI-aided approach, respectively.

In comparison with the manual and AI-aided approaches, the automated approach yielded significantly higher values for the LAM thickness, OIM thickness, and the IPPL. The higher automated muscle thickness could be explained by the fact that the segmentations around the muscle borders were slightly larger in some cases. In addition, variations in muscle diameters may be attributed to the computer’s capability to accurately compute the maximum diameter, while variations in angle or slice positioning during manual measurements by radiologists may affect the measured diameter. In future studies, it may be beneficial to present the computed thickness to radiologists, enabling them to verify the measurements and potentially minimize the need for manual measurements. The automated IPPL measurements also demonstrated higher values compared to the manual and AI-aided measurements, often resulting in an overestimation of the protrusion. Consequently, manual assessments of DL-generated segmentations remain important, especially in cases where segmentation errors may arise.

The automated sagittal and coronal MUL measurements were not statistically different from the AI-aided measurements. The automated sagittal MUL measurements were 14.32 mm, the AI-aided measurements were 14.04 mm, and the manual measurements were 14.94 mm. The larger manual MUL measurements can be attributed to the different positioning of the superior point of the membranous urethra. During the first session, two radiologists defined the superior point in some patients differently, placing it more superiorly where the urethra was still visible in the caudal portion of the prostate apex. This superior point may align more closely with the surgical section during RARP, where urologists aim to preserve the membranous urethra for the vesicourethral anastomosis [14], except in cases involving apical tumors. In the second session, the superior point was consistently positioned at the inferior border of the prostate due to the DL-generated segmentations, which contributed to increased consistency as indicated by the AI-aided measurements. This definition of the MUL measurement was comparable to previous studies [3, 6, 7].

The MUL measurements in our study were comparable with Kim et al. [3], who reported mean values of 14.6 mm in the coronal direction and 14.2 mm in the sagittal direction. Veerman et al. [7] found higher MUL measurements of 17 mm, which could potentially be attributed to the inclusion of patients with higher MUL values since the introduction of a risk prediction model of urinary incontinence in their institution [15]. LAM measurements were comparable to the findings of Sadahira et al. [16] but higher than the measurements reported by Muñoz-Calahorro et al. [6], which ranged between 4.22 and 6.87 mm. These differences in measurements could be attributed to variances in anatomical definition, as Muñoz-Calahorro et al. [6] measured the thickness before the insertion of the muscle puborectalis fibers, while we measured the largest thickness regardless of its location.

More consistent pelvic measurements are especially clinically relevant when including them in urinary incontinence prediction models for patients undergoing RARP. Multiple studies have already created prediction models for continence outcome after RARP and incorporated patient-related variables and MRI-based variables like the MUL [10, 17,18,19]. Standardized measurements are essential since the continence prediction tool described in Tillier et al. [10] demonstrates that a 1-mm increase in the MUL corresponds to an average increase of 8.1 percentage points in the chance of continence recovery after six months of RARP [7]. These findings can influence the treatment choices of PCa patients. Therefore, standardizing prostate MRI measurements using DL is an unmet need for continence prediction models. Additionally, DL-generated segmentations provide an opportunity to assess additional geometric variables, including shape-based features, surface areas, and volume-based features.

This study had some limitations. First, not all training cases were verified by three radiologists but a subset of 70 patients was evaluated and the feedback of the three different radiologists was used. Second, the manual assessment of the pelvic measurements was performed in sagittal and coronal plane. It was not possible to visualize the axial, coronal, and sagittal planes simultaneously. The visualization of the prostate in three directions would have enhanced the ease of evaluating MUL measurements. Third, external validation of our DL model and automated measurements in other centers is essential because only a part of the T2-weighted images adhered to the PI-RADSv2 MRI acquisition parameter guideline [20]. We believe that our DL model has the potential for generalizability, as it has been trained using data from five regional hospitals with various magnetic field strengths, vendors, and sequence protocols. Additional training cases with variations in anatomy, such as more patients with IPP and patients with hip prostheses, could contribute to improving the generalization ability of our DL model. Finally, the clinical impact of AI-aided assessment of pelvic parameters could not be retrospectively assessed. Future studies are needed to evaluate the relationship between objective pelvic measurements and postoperative outcomes.

In conclusion, our study demonstrates the potential of deep learning (DL)-generated segmentations to improve interobserver agreement of pelvic measurements among radiologists. The membranous urethral length (MUL) measurement improved from substantial agreement to almost perfect agreement among radiologists with DL-generated segmentations of pelvic structures. Standardized pelvic measurements can be incorporated into urinary incontinence prediction models for patients undergoing RARP.

Availability of data and materials

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.



Artificial intelligence


Confidence interval


Deep learning


Dice similarity coefficient


Intraclass correlation coefficient


Intravesical prostatic protrusion length


Interquartile range


Levator ani muscle


Magnetic resonance imaging


Mean surface distances


Membranous urethral length


Obturator internus muscle


Prostate cancer


Robot-assisted radical prostatectomy


  1. Lane JA, Donovan JL, Young GJ et al (2022) Functional and quality of life outcomes of localised prostate cancer treatments (Prostate Testing for Cancer and Treatment [ProtecT] study). BJU Int 130:370–380.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  2. van Dijk-de Haan MC, Boellaard TN, Tissier R et al (2022) Value of different magnetic resonance imaging-based measurements of anatomical structures on preoperative prostate imaging in predicting urinary continence after radical prostatectomy in men with prostate cancer: a systematic review and meta-analysis. Eur Urol Focus 8:1211–1225.

    Article  PubMed  Google Scholar 

  3. Kim LHC, Patel A, Kinsella N, Sharabiani MTA, Ap Dafydd D, Cahill D (2020) Association between preoperative magnetic resonance imaging–based urethral parameters and continence recovery following robot-assisted radical prostatectomy. Eur Urol Focus 6:1013–1020.

    Article  PubMed  Google Scholar 

  4. Lardas M, Grivas N, Debray TPA et al (2022) Patient- and tumour-related prognostic factors for urinary incontinence after radical prostatectomy for nonmetastatic prostate cancer: a systematic review and meta-analysis. Eur Urol Focus 8:674–689.

    Article  PubMed  Google Scholar 

  5. Lamberg H, Shankar PR, Singh K et al (2022) Preoperative prostate MRI predictors of urinary continence following radical prostatectomy. Radiology 303:99–109.

    Article  PubMed  Google Scholar 

  6. Muñoz-Calahorro C, Parada-Blázquez MJ, García-Sánchez C, López-Arellano L, Vizcaíno-Velázquez P, Medina-López RA (2023) Inter-observer variability in male pelvic-floor MRI measurements that might predict post-prostatectomy incontinence. World J Urol 41:1147–1155.

    Article  PubMed  Google Scholar 

  7. Veerman H, Hagens MJ, Hoeks CM et al (2022) A standardized method to measure the membranous urethral length (MUL) on MRI of the prostate with high inter- and intra-observer agreement. Eur Radiol 33:3295–3302.

    Article  PubMed  Google Scholar 

  8. Boellaard TN, Hagens MJ, Veerman H et al (2023) Prostate MRI for improving personalized risk prediction of incontinence and surgical planning: the role of membranous urethral length measurements and the use of 3D models. Life 13:830.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Liu G, Pan S, Zhao R et al (2023) The added value of AI-based computer-aided diagnosis in classification of cancer at prostate MRI. Eur Radiol 33:5118–5130.

    Article  PubMed  Google Scholar 

  10. Tillier CN, Vromans RD, Boekhout AH et al (2021) Individual risk prediction of urinary incontinence after prostatectomy and impact on treatment choice in patients with localized prostate cancer. Neurourol Urodyn 40:1550–1558.

    Article  PubMed  Google Scholar 

  11. Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier-Hein KH (2021) nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods 18:203–211.

    Article  CAS  PubMed  Google Scholar 

  12. van den Berg I, Savenije MHF, Teunissen FR et al (2023) Deep learning for automated contouring of neurovascular structures on magnetic resonance imaging for prostate cancer patients. Phys Imaging Radiat Oncol 26:100453.

    Article  PubMed  PubMed Central  Google Scholar 

  13. Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33:159–174.

    Article  CAS  PubMed  Google Scholar 

  14. Schlomm T, Heinzer H, Steuber T et al (2011) Full functional-length urethral sphincter preservation during radical prostatectomy. Eur Urol 60:320–329.

    Article  PubMed  Google Scholar 

  15. Grivas N, van der Roest R, Schouten D et al (2018) Quantitative assessment of fascia preservation improves the prediction of membranous urethral length and inner levator distance on continence outcome after robot-assisted radical prostatectomy. Neurourol Urodyn 37:417–425.

    Article  PubMed  Google Scholar 

  16. Sadahira T, Mitsui Y, Araki M et al (2019) Pelvic magnetic resonance imaging parameters predict urinary incontinence after robot-assisted radical prostatectomy. Low Urin Tract Symptoms. 11:122–126.

    Article  PubMed  Google Scholar 

  17. Gu Z, Zheng Z, Zhang W et al (2023) The development and assessment of a predicting nomogram for the recovery of immediate urinary continence following laparoscopic radical prostatectomy. Front Surg 9:1071093.

    Article  PubMed  PubMed Central  Google Scholar 

  18. Zhang F, Chu H, Hao Y et al (2023) Preoperative predictive model of early urinary continence recovery after laparoscopic radical prostatectomy. World J Urol 41:59–65.

    Article  CAS  PubMed  Google Scholar 

  19. Onishi T, Sekito S, Kajiwara S, Shibahara T (2022) Simple and reliable predictive factor for early recovery of urinary continence after non-nerve-sparing robot-assisted laparoscopic radical prostatectomy. J Robot Surg 17:1077–1083.

    Article  PubMed  Google Scholar 

  20. Cuocolo R, Stanzione A, Ponsiglione A et al (2019) Prostate MRI technical parameters standardization: a systematic review on adherence to PI-RADSv2 acquisition protocol. Eur J Radiol 120:108662.

    Article  PubMed  Google Scholar 

Download references


The authors state that this work has not received any funding.

Author information

Authors and Affiliations



IB trained and evaluated the deep learning models. IB and RN developed the post-processing steps. FW, EH, and CN performed a clinical evaluation of the deep learning models. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Ingeborg van den Berg.

Ethics declarations

Ethics approval and consent to participate

This study received approval from the institutional review and ethics board of the University Medical Center Utrecht (19–692/M) and patients provided informed consent for sharing their data within the Utrecht Prostate Cohort (UPC) study (NCT04228211).

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1:

 Supplementary Figure 1. Bland-Altman plots of measured membranous urethral length (MUL) in the coronal direction for manual assessment and AI-aided assessment. Supplementary Figure 2. Bland-Altman plots of measured intravesical prostatic protrusion length (IPPL) in the sagittal direction for manual assessment and AI-aided assessment. Supplementary Figure 3. Bland-Altman plots of measured obturator internus muscle (OIM) thickness in the coronal direction for manual assessment and AI-aided assessment. Supplementary Figure 4. Bland-Altman plots of measured levator ani muscle (LAM) thickness in the coronal direction for manual assessment and AI-aided assessment.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

van den Berg, I., Spaans, R.N., Wessels, F.J. et al. Automated pelvic MRI measurements associated with urinary incontinence for prostate cancer patients undergoing radical prostatectomy. Eur Radiol Exp 8, 1 (2024).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: