UP-158 Development of a machine learning algorithm for Peyronie’s disease curvature assessment
Thursday June 27, 2019 from
TBD
Presenter

Luke Witherspoon, Canada

Fellow

Urology

The University of British Columbia

Abstract

Development of a machine learning algorithm for Peyronie’s disease curvature assessment

Luke Witherspoon1,2, Reza Soltani4, James Gleave1, Faraz Hach1,3, Ryan Flannigan1,5.

1Department of Urologic Sciences, University of British Columbia, Vancouver, BC, Canada; 2Division of Urology, University of Ottawa, Ottawa, ON, Canada; 3Vancouver Prostate Centre, University of British Columbia, Vancouver, BC, Canada; 4Bioinformatics Department, University of British Columbia, Vancouver, BC, Canada; 5Department of Urology, Weill Cornell Medicine, New York, NY, United States

Introduction: Although Peyronie’s disease prevalence estimates suggest that 11-15% of men suffer from this disease, only 1% seek evaluation. When patients do seek help, tools for penile curvature assessment are lacking. We present a tool for patient and clinician led assessment of Peyronie’s disease using a neural network system capable of plaque assessment of photographic images.

Methods: Men underdoing penile curvature assessment in a sexual medicine clinic were recruited to enroll in this study. Images were taken of the erect penis from the left, right and top sides using an Apple iPad™.  For initial model training 150 images were used from the open access Not Suitable for Work (NSWF) repository (https://github.com/EBazarov/nsfw_data_source_urls/tree/master/raw_data) and labelled manually for object detection. Subsequently model training was conducted with a residual neural network (ResNet), a neural network-based framework used for image recognition, to identify male genitals. To assess basic curvature a mathematical formulation was created to calculate the penile curvature after fitting linear lines from genital tip to curvature point and from curvature point to the base of the penis.

Results: Initial training of the image analysis model has allowed for identification of the presence of a curvature of a penis, with ongoing training for curvature quantification (Figure 1). We have trained the model to remove background images, isolating the genitals (Figure 1) to aid in patient privacy.

Conclusions: This study provides a proof of concept that machine learning can augment the assessment of patients with Peyronie’s disease. With demonstration of curvature assessment via photographic images this will allow patients to perform an initial curvature assessment at home, improving motivation for patients to seek help.  Furthermore, this technology can augment clinician led curvature assessment, providing an objective and easy means for documenting disease severity.


© 2022 CUA 74th Annual Meeting