Optimizing Anterior Urethral Stricture Assessment: Leveraging AI-Assisted Three-Dimensional Sonourethrography in Clinical Practice

doi:10.21203/rs.3.rs-4356988/v1

Download PDF

Research Article

Optimizing Anterior Urethral Stricture Assessment: Leveraging AI-Assisted Three-Dimensional Sonourethrography in Clinical Practice

https://doi.org/10.21203/rs.3.rs-4356988/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 01 Jul, 2024

Read the published version in International Urology and Nephrology →

Version 1

posted

You are reading this latest preprint version

Purpose

This investigation sought to validate the clinical precision and practical applicability of AI-enhanced three-dimensional sonographic imaging for the identification of anterior urethral stricture.

Methods

The study enrolled 63 male patients with diagnosed anterior urethral strictures alongside 10 healthy volunteers to serve as controls. The imaging protocol utilized a high-frequency 3D ultrasound system combined with a linear stepper motor, which enabled precise and rapid image acquisition. For image analysis, an advanced AI-based segmentation process using a modified U-net algorithm was implemented to perform real-time, high-resolution segmentation and three-dimensional reconstruction of the urethra. A comparative analysis was performed against the surgically measured stricture lengths. Spearman’s correlation analysis was executed to assess the findings.

Results

The AI model completed the entire processing sequence, encompassing recognition, segmentation, and reconstruction, within approximately 5 minutes. The mean intraoperative length of urethral stricture was determined to be 14.452 ± 8.385 mm. Notably, the mean lengths of the urethral strictures reconstructed by manual and AI models were 13.092 ± 7.463 mm and 13.419 ± 7.216 mm, respectively. Interestingly, no statistically significant disparity in urethral stricture length between manually reconstructed and AI-reconstructed images was observed. Spearman’s correlation analysis underscored a more robust association of AI-reconstructed images with intraoperative urethral stricture length than manually reconstructed 3D images (0.869811 vs. 0.8201756). Furthermore, AI-reconstructed images provided detailed views of the corpus spongiosum fibrosis from multiple perspectives.

Conclusions

The research heralds the inception of an innovative, efficient AI-driven sonographic approach for three-dimensional visualization of urethral strictures, substantiating its viability and superiority in clinical application

Urethral Stricture

Sonourethrography

Artificial intelligence

Digital image

Three-dimensional image

In the intricate landscape of urethral stricture management, the meticulous diagnosis and precise assessment of stricture intricacies serve as pivotal guides for surgical decision-making¹. Imaging, positioned at the core of this diagnostic endeavor, assumes a paramount role in unraveling the nuanced details of the stricture. For more than a century, retrograde urethrogram (RUG) and voiding cystourethrography (VCUG) have stood as stalwart pillars, universally recognized as the gold standard methodologies for evaluating urethral strictures. Esteemed for their cost-effectiveness, extensive availability, and commendable accuracy, these techniques have solidified their status in clinical practice². Moreover, advanced examinations, such as sonourethrography (SUG) and magnetic resonance urethrography (MRU), have also been applied to diagnose urethral stricture ^{3 4}.

Compared to the traditional RUG & VCUG, SUG has many advantages for diagnosing anterior and bulbar urethral strictures, such as non-radioactive hazard, less invasive, low cost, and real-time monitoring. Moreover, SUG strongly correlates with intraoperative findings and provides more details about periurethral fibrosis ⁴. In our preceding investigation, we pioneered developing an innovative three-dimensional computerized model grounded in sonourethrography (SUG) to enhance the examination of urethral strictures. This novel model offers a unique vantage point, enabling clear visualization and comprehensive comprehension of stricture details and periurethral fibrosis. Its user-friendly interface ensures accessibility, rendering complex anatomical insights readily understandable to a wide array of stakeholders⁵.

However, our technique encounters notable limitations, primarily revolving around the time-intensive nature of image reconstruction, compounded by the manual segmentation required to delineate the region of interest. Typically, this process demands nearly a full day for technicians to meticulously reconfigure images to ensure comprehensive visualization of all stricture details. Addressing this inherent deficiency has emerged as a pivotal endeavor in realizing the clinical applicability of our methodology. Thus, we focused on leveraging artificial intelligence, particularly convolutional neural networks, to streamline the image reconstruction process and substantially reduce the time investment. This innovative approach holds promise in facilitating the realization of computer-aided corpus spongiosum segmentation (CACSS). In the current study, we have extended our efforts by refining an ultrasonic, AI-automated three-dimensional reconstructive model tailored for assessing the status of the anterior urethra.

Patient summary

Between January 2019 and June 2023, 63 male patients with an anterior urethral stricture and 10 healthy male volunteers were enrolled in this study. The ages of the participants ranged from 25 to 63 years, and the mean age was 35.35 years. The initial diagnosis of stricture was made based on the patient’s history and the findings of RUG and uroflowmetry. The exclusion criteria were active urinary tract infection, bladder outlet obstruction, or urinary retention caused by conditions other than urethral stricture. Institutional review board approval was obtained from the Shanghai Sixth People’s Hospital. (No. YS-2019-45). All the work was performed in accordance with the 1964 Declaration of Helsinki, and informed consent was obtained from all the subjects.

Ultrasound equipment modification

A high sample rating linear motion 3D ultrasound system was used in this study. The composition of this system is shown in Fig. 1. In this system, we chose a high-frequency transducer as the ultrasound probe, with a median frequency of 8.5 MHz (6 ~ 14 MHz). More importantly, we designed a linear stepper motor for image collection (Fig. 1C, D). With this motor, the probe can linearly move, and the image collection rate can reach 100 frames/second. The final spatial resolution was greater than 0.1 mm, which was used for subsequent 3D spatial analysis.

Ultrasonography

The process of ultrasonography was described previousl⁵. Briefly, sterile water was injected using a 20-ml syringe, and care was taken not to inject any air bubbles. The penis was cranially extended over the pubic bone, and the top of the syringe blocked the meatus. Water was infused into the bladder if the patient had a suprapubic catheter. Then, the catheter was clamped, and the patient was asked to urinate. The transducer with the newly designed linear motor was placed on the ventral surface of the penis. Longitudinal and transverse multisection scans were performed while the urethral lumen maintained a steady width during constant-speed injection. Images were subsequently uploaded to the AI system for further imaging segmentation and 3D model reconstruction.

AI image segmentation and 3D sonourethrography reconstruction

3D reconstruction and rendering of the penis can be divided into two stages. The first stage is an artificial intelligence (AI)-based image segmentation step on individual 2D images that identifies and separates the anatomical structures of interest. Because segmentation is performed for each image frame, the result is a 3D volume of the anatomical structures we wish to view. The second stage is a cinematic rendering of the ultrasound data that provides a realistic and comprehensive view of the penile anatomy together with the pathology of interest.

AI-based image segmentation is based on a modified U-net (Fig. 1E) that segments pixels into four classes: urethral lumen, corpus spongiosum, periurethral fibrosis, and all other tissue.⁶ The training data for the four classes were labeled by a urologist with ten years of experience. The images were resized to a grayscale 256 × 256 input size for network input.

For 3D rendering, voxels identified by U-net as the urethral lumen were set to 0 to remove reflection or reverberation artifacts. This difference may alternatively be interpreted as setting these voxels to be fully transparent. The voxels corresponding to the periurethral fibrosis were assigned an opacity function and colormap that was different from all the other remaining voxels. Thus, we used two sets of opacity functions and colormaps to visually differentiate the fibrosis voxels. We use a raycaster enhanced with physically based light transport for the final rendering⁷. The intraoperative length of the stricture was also recorded and was used to compare the information for the AI-reconstructed anterior urethra.

Statistical analysis

Continuous variables are reported as the mean and range. Spearman’s correlation analysis was used to evaluate the correlation between intraoperative stricture measurements and AI 3D imaging data or manually reconstructed 3D imaging. A p value < 0.05 was considered to indicate statistical significance. The statistical data were analyzed using STATISTICA (Statistica 12.0, StatSoft, Inc., USA).

The etiologies of anterior and bulbar strictures included intragenic injury in 11 patients (11/63), inflammatory injury in 17 patients (17/63), and straddle injury in 35 patients (35/63). The stricture in 42 patients was located in the penile urethra. Twenty-one patients had strictures in the bulbar urethra.

All patients successfully underwent SUG without complications. For the segmentation of the urethral image, an experienced urologist spent totally 10 to 16 hours to segment the urethra and corpus spongiosum. The duration of this process depended on the length of the targeted urethra. It took AI less than 5 minutes to complete the process of recognition and segmentation.

The 3D reconstructed urethral images obtained via manual processing and the Unet AI model process are shown in (Fig. 2A-D) The mean intraoperative length of the urethral stricture was 14.452 ± 8.385 mm. The mean lengths of the urethral strictures reconstructed manually and those reconstructed by the AI model were 13.092 ± 7.463 mm and 13.419 ± 7.216 mm, respectively. There was no statistically significant difference in urethral stricture length between the manual and AI-generated models. Furthermore, Spearman’s correlation analysis demonstrated that the 3D image reconstructed from the AI model had a stronger association with the intraoperative urethral stricture length than that in manually reconstructed 3D image (0.869811 vs. 0.8201756) (Fig. 2E,F).

Moreover, the AI model-generated urethral image could directly reveal the distribution of corpus spongious fibrosis. By using different colors and angle views, it is easy for urologists to evaluate the severity of strictures and fibrosis(Figure 3A,B). Interestingly, the AI model produced a special 3D image in the sagittal plane, which has never been previously observed in traditional images. This approach provided a more direct view of the urethral stricture and fibrosis in a single image（Figure 3C-F）.

In the conducted research, we have markedly enhanced the protocol for anterior urethral stricture visualization using 3D sonourethrography, augmented by an advanced AI model. Implementing this innovative technique has led to a significant reduction in the duration of the imaging process. Furthermore, the automated AI image segmentation demonstrates equivalence in quality to meticulous manual segmentation processes. There was a noteworthy concordance between the AI-generated 3D images of urethral strictures and the actual intraoperative findings, advocating for integrating this technique into contemporary clinical practice. This novel approach supersedes traditional 2D SUG by producing intuitive, readily interpretable 3D images of the urethra. The AI model ensures the reproducibility of this technique, making it suitable for diverse clinical environments, including immediate use in the operating room.

The meticulous assessment of urethral strictures is paramount, serving as a cornerstone for surgeons in crafting a nuanced preoperative strategy that encompasses the selection of an appropriate surgical intervention⁸. Historically, retrograde urethrography (RUG) and voiding cystourethrography (VCUG) have been integral to diagnosing urethral stricture. Recent developments, however, have seen the adoption of ultrasonographic imaging and MRI as supplementary diagnostic tools for a spectrum of urethral disorders⁹. MRI urethrography provides enhanced accuracy in delineating strictures located in the proximal urethra. Yet, for strictures within the anterior urethra, no significant diagnostic discrepancy is observed between MR urethrography and sonourethrography.². On the contrary, the high cost, relatively narrow indication, and longer learning curve of radiologists are inherent limitations of MRU.

Sonourethrography (SUG) stands out as a superior, cost-effective technique for assessing anterior urethral conditions, noted for its simplicity, reproducibility, and accuracy.¹⁰. Despite these compelling advantages, its widespread clinical adoption has been limited. This limitation can be attributed to several factors: the shortfall of in-depth ultrasonography training among urology professionals, the necessity for adept technicians specialized in SUG, and a prevalent unfamiliarity with SUG image interpretation within the clinical community. ⁹ Addressing these challenges, our study introduces a novel methodological approach that enhances the SUG technique. This advancement facilitates intuitive, three-dimensional visualizations of the urethra, streamlining the interpretation process and making it more accessible to healthcare providers. This intuitive interpretation is crucial, particularly in diagnosing anterior urethral strictures, thus emphasizing the need to refine SUG imaging methods further to cement its place as an indispensable tool in urological diagnostics.

In recent studies, contrast-enhanced urosonography (CEUS) was introduced for the evaluation of urethral strictures^11–13. Benson et al. ¹³ suggested that, compared with grayscale SUG, CEUS SUG might be more effective at delineating the urethra and a stricture. On CEUS-SUG, narrow caliber lumens are easier to detect. The authors utilized CEUS-SUG to evaluate the degree of postoperative urethral patency with high sensitivity, specificity, and accuracy compared to cystoscopy. Our study used normal saline instead of traditional contrast media for sonourethrography (SUG). We found that it provided sufficient contrast for effective AI segmentation of the urethral lumen and corpus spongiosum. This approach not only aligns with actual intraoperative observations, affirming saline's adequacy for detailed SUG image analysis but also promises a substantial reduction in costs. The potential for decreased expenses and increased simplicity in image interpretation positions saline-based SUG as an appealing alternative for widespread clinical adoption.

Shear wave elastography (SWE)¹³ is another technique that can help to localize and quantify tissue stiffness in the corpus spongiosum. SWE can guide surgical treatment and predict stricture recurrence. With our reconstructed 3D urethral image, we demonstrated the distribution of fibrosis around the urethra. However, our technique could not observe stiffness in the fibrosis model. Using SWE, we can further evaluate the severity of spongy tissue invasion. The results can help urologists accurately classify urethral strictures¹⁴. Regrettably, the seamless fusion of SWE imaging with three-dimensional urethral reconstruction remains elusive, utilizing contemporary methodologies. This challenge presents an opportunity for refinement and should be a focal point for future investigative pursuits

The key innovation in this study is that artificial intelligence (AI) supersedes manual analysis in image processing. Recently, the application of AI within the field of imaging diagnostics has led to remarkable achievements across a range of subfields, such as in the diagnosis of prostate, lung, and skin cancer ¹⁵. Arsenescu et al.¹⁶ recently used the MultiResUNet model for 3D ultrasound reconstructions of the carotid artery and thyroid gland. A qualitative evaluation compared the US results with the CT scanning results. The overall scores for automated segmentation using MultiResUNet are ideal. This study proved the feasibility of using an AI model for 3D ultrasound images. Kim et al.¹⁷ used a convolutional neural network (CNN)-based machine learning algorithm to characterize RUG images. Their results showed that this algorithm could correctly characterize 88.5% of the images. Our research is a testament to AI's transformative power in medical imaging, significantly expediting the segmentation process without sacrificing precision. Leveraging a modified Unet model, akin to Arsenescu's methodology, we've achieved notable accuracy in image recognition and segmentation—comparable to manual techniques. These advancements suggest that our AI-enhanced approach is efficient and poised for integration into clinical settings, heralding a new era of diagnostic capability."

Another development in our study is that we have employed a state-of-the-art linear stepper motor to facilitate image acquisition. This technological advancement overcomes the constraints imposed by the traditional probe's limited width. Previously, our capacity to reconstruct the anterior urethra was restricted to segments ranging from 0.5 to 4 cm. Introducing this sophisticated apparatus marks a significant leap in our imaging capabilities, enabling comprehensive visualization and, consequently, more detailed analysis.⁵. The newly developed equipment allows us to reconstruct longer 3D anterior urethral images simultaneously. Previously, researchers used the technique of rapid imaging stitching to overcome the limitation of the small field of view (FOV) in ultrasound imaging¹⁸. However, obvious bias from image stitching cannot be avoided. It will take a long time to perform urethral 3D imaging reconstruction for two or more scans; therefore, the qualification of the final image cannot be accepted. Since an increasing number of studies have used linear stepper motors for ultrasound image collection, we designed our specific stepper in our study^19,20. With this equipment, more frames can be collected stably and efficiently than ever, achieving high-quality, large FOVs for 3D reconstruction.

There are still other limitations in our study. SUG is also limited in its ability to define posterior urethral strictures at present ²¹. Current methodologies fail to provide a unified imaging solution for anterior and posterior urethral evaluation. Our proposed technique, thus far, demonstrates an enhanced capacity for anterior urethral assessment. It is worth noting that for posterior urethral strictures. Another limitation of this study is its single-center design, which may not adequately represent broader clinical scenarios. Future endeavors aim to expand the validation of this technique across multiple institutions, thereby enhancing its generalizability and clinical applicability.

This study introduces an innovative urethral 3D sonography technique enhanced by artificial intelligence (AI). This pioneering approach streamlines the process of image reconstruction and yields superior-quality 3D reconstructed urethral images. Our findings propose that this advanced methodology holds promise for both clinical assessments and intraoperative evaluations of anterior urethral disorders.

Author Contribution

Conceived and planned the study: Chao Feng, Lei Chen; principal investigator: Lei Chen,Yue-min Xu; sample preparation:Ying-Long Sa;statistical analysis and interpretation of the results: Qi-jie Lu; writing the manuscript:Jin-Dong Xue,Hui-Quan Shu.All Authors performed or supervised the surgeries in both arms. All Authors provided critical feedback and helped shape the research, analysis, and manuscriptAll Authors read and approved the final manuscript.

Statements and Declarations

There are non-financial interests that are directly or indirectly related to the work submitted for publication

Harris D, Zhou C, Girardot J, Kidron A, Gupta S, Cavalcanti AG, Bittencourt LK.(2023) Imaging in urethral stricture disease: an educational review of current techniques with a focus on MRI. Abdom Radiol (NY) 48:1062-1078. doi: 10.1007/s00261-022-03761-5
Mikolaj F, Karolina M, Oliwia K, Jakub K, Adam K, Mariusz B, Patrycja N, Marcin M. (2021) Retrograde urethrography, sonouretrography and magnetic resonance urethrography in evaluation of male urethral strictures. Should the novel methods become the new standard in radiological diagnosis of urethral stricture disease? Int Urol Nephrol. 53:2423-2435. doi: 10.1007/s11255-021-02994-5
Morey AF, McAninch JW. (1996) Ultrasound evaluation of the male urethra for assessment of urethral stricture. J Clin Ultrasound.24:473-479. doi: 10.1002/(SICI)1097-0096(199610)24:8<473::AID-JCU7>3.0.CO;2-H
Revels JW, Wang SS, Weaver JS, Foreman JR, Gallegos MA, Thompson WM, Katz D, Moshiri M.(2022) A multimodality review of male urethral imaging: pearls and pitfalls with an update on urethral stricture treatment. Br J Radiol. 95:20211034. doi: 10.1259/bjr.20211034
Chen L, Feng C, Lv XG, Fan HH, Joshi P, Barbagli G, Zhao YM, Xu YM, Li Z, Hu B .(2018) Three-Dimensional Computerized Model Based on the Sonourethrogram: A Novel Technique to Evaluate Anterior Urethral Stricture. J Urol. 199:568-575. doi: 10.1016/j.juro.2017.08.098
O. Ronneberger PFaTB. (2015)U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention. 234-241.
Kroes T, Post FH, Botha CP. (2012)Exposure render: an interactive photo-realistic volume rendering framework. PLoS One. 7:e38586. doi: 10.1371/journal.pone.0038586
Sussman RD, Kozirovsky M, Telegrafi S, Peyronnet B, Palmerola R, Smilen S, Pape DM, Rosenblum N, Nitti VW, Brucker BM.（2020） Gel-Infused Translabial Ultrasound in the Evaluation of Female Urethral Stricture. Female Pelvic Med Reconstr Surg. 26:737-741. doi: 10.1097/SPV.0000000000000699
Theisen KM, Kadow BT, Rusilko PJ. （2016）Three-Dimensional Imaging of Urethral Stricture Disease and Urethral Pathology for Operative Planning. Curr Urol Rep. 17:54. doi: 10.1007/s11934-016-0616-0
Bryk DJ, Khurana K, Yamaguchi Y, Kozirovsky M, Telegrafi S, Zhao LC. （2016）Outpatient Ultrasound Urethrogram for Assessment of Anterior Urethral Stricture: Early Experience. Urology. 93:203-207. doi: 10.1016/j.urology.2016.03.016
Fischer KM, Bowen DK, Kovell RC, Back SJ, Darge K, Weiss DA. （2020）Intraoperative contrast enhanced sonourethrography to characterize urethral stricture in a pediatric patient. Urol Case Rep. 32:101223. doi: 10.1016/j.eucr.2020.101223
Chung PH, Leong JY, Machado P, Wessner CE, Trabulsi EJ, Halpern EJ, Eisenbrey JR, Forsberg F.（2022） Contrast-Enhanced Ultrasound and Shear Wave Elastography: Novel Methods for the Evaluation of Urethral Stricture Disease. J Urol. 207:152-160. doi: 10.1097/JU.0000000000002146
Benson CR, Brandes SB. （2022）Contrast-Enhanced Ultrasound and Shear Wave Elastography: Novel Methods for the Evaluation of Urethral Stricture Disease. Letter. J Urol. 208:563-564. doi: 10.1097/JU.0000000000002786
Chiou RK, Anderson JC, Tran T, Patterson RH, Wobig R, Taylor RJ （1996）. Evaluation of urethral strictures and associated abnormalities using high-resolution and color Doppler ultrasound. Urology. 47:102-107. doi: 10.1016/s0090-4295(99)80391-4
Li G, Tian H, Wu H, Huang Z, Yang K, Li J, Luo Y, Shi S, Cui C, Xu J.（2023） Artificial intelligence for non-mass breast lesions detection and classification on ultrasound images: a comparative study. BMC Med Inform Decis Mak.23:174. doi: 10.1186/s12911-023-02277-2
Arsenescu T, Chifor R, Marita T, Santoma A, Lebovici A, Duma D, Vacaras V, Badea AF（2023）. 3D Ultrasound Reconstructions of the Carotid Artery and Thyroid Gland Using Artificial-Intelligence-Based Automatic Segmentation-Qualitative and Quantitative Evaluation of the Segmentation Results via Comparison with CT Angiography. Sensors (Basel).;23. doi: 10.3390/s23052806
Kim JK, McCammon K, Robey C, Castillo M, Gomez O, Pua PJL, Pile F, See Mt, Rickard M, Lorenzo AJ.（2022） Identifying urethral strictures using machine learning: a proof-of-concept evaluation of convolutional neural network model. World J Urol. 40:3107-3111. doi: 10.1007/s00345-022-04199-6
Chang RF, Chang-Chien KC, Takada E, Huang CS, Chou YH, Kuo CM, Chen JH.（2010） Rapid image stitching and computer-aided detection for multipass automated breast ultrasound. Med Phys. 37:2063-2073. doi: 10.1118/1.3377775
Mozaffarzadeh M, Moore C, Golmoghani EB, Mantri Y, Hariri A, Jorns A, Fu L, Verweij MD, Orooji M, de Jong N. (2021) Motion-compensated noninvasive periodontal health monitoring using handheld and motor-based photoacoustic-ultrasound imaging systems. Biomed Opt Express. 12:1543-1558. doi: 10.1364/BOE.417345
Sangha GS, Hu B, Li G, Fox SE, Sholl AB, Brown JQ, Goergen CJ. (2022)Assessment of photoacoustic tomography contrast for breast tissue imaging using 3D correlative virtual histology. Sci Rep. 12:2532. doi: 10.1038/s41598-022-06501-3
Ravikumar BR, Tejus C, Madappa KM, Prashant D, Dhayanand GS.(2015) A comparative study of ascending urethrogram and sono-urethrogram in the evaluation of stricture urethra. Int Braz J Urol. 41:388-392. doi: 10.1590/S1677-5538.IBJU.2015.02.30

No competing interests reported.

Download PDF

Journal Publication

published 01 Jul, 2024

Read the published version in International Urology and Nephrology →

Version 1

posted

You are reading this latest preprint version

Optimizing Anterior Urethral Stricture Assessment: Leveraging AI-Assisted Three-Dimensional Sonourethrography in Clinical Practice

Status:

Journal Publication

Version 1

Abstract

Purpose

Methods

Results

Conclusions

Figures

Introduction

Materials and Methods

Patient summary

Ultrasound equipment modification

Ultrasonography

AI image segmentation and 3D sonourethrography reconstruction

Statistical analysis

Results

Discussion

Conclusion

Declarations

Author Contribution

Statements and Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1