PenoMeter: a machine learning and algorithmic tool to advance Peyronie's disease assessment.

Abstract

Background: Peyronie's disease curvature assessment is a critical step for patient assessment; however, tools for objective, unbiased, and reproducible quantification are currently limited.

Aim: To develop an automated computational tool to identify the penis from a 2D image and to accurately and reproducibly measure the degree of angulation.

Methods: We developed PenoMeter using instance segmentation to identify penile anatomical components, key point detection to identify shaft corners, geometric calculations to locate and measure the angulation of the point of maximal curvature. We trained our model on training datasets and evaluated the PenoMeter using a separate dataset of digital penile images.

Outcomes: The PenoMeter is an artificial intelligence-powered assistive diagnostic toolkit that can automatically assess the curvature angle of penile 2D images that holds potential for healthcare practitioners to use in assistance for PD assessments.

Results: The PenoMeter's reported angulation, relative to the mean angulation reported by three subspecialized urologists, falls within their range of variability in 57 out of 66 cases (86%) and outside their range of variability in 9 out of 66 cases (14%) of digital images. The PenoMeter demonstrated no intra-observer variance (0°) in repeated measures over time compared to the three subspecialized urologists who demonstrated intra-observer variability between by 3.8° to 7.8°.

Clinical and translational implications: The PenoMeter can be utilized for initial PD assessment and tracking treatment outcomes in time-series data for both clinical and research contexts.

Strengths and limitations: Strengths of the PenoMeter include unbiased and objective quantification of penile curvature. Furthermore, it demonstrates no intra-observer variability, making it appealing for evaluating time-series digital images. Limitations of the PenoMeter include the lack of a measure of confidence for curvature assessment. Detection and measurement of other forms of PD deformities such as indentations, hourglass deformities, torque and distal tapering require further development. Finally, accurate curvature quantification is reliant on reproducibly acquiring accurate digital images and an accurate and consistent assessment of penile rigidity; therefore, a well-defined process for image acquisition and clinician assessment of penile rigidity immediately prior to digital photo capture would be required to enhance accuracy of obtaining a representatively accurate image for processing.

Conclusions: The PenoMeter's performance in penile curvature assessment of digital photos are objective, accurate and reproducible, and therefore carries potential to assist clinicians' initial PD assessments and treatment outcome tracking. However, the PenoMeter is not currently positioned to replace the current gold-standard in-office assessment.