Standardization of penile angle estimation with a semi-automated algorithm

Penile curvature (PC) refers to an abnormal bending of the main longitudinal axis of the penis. It is frequently associated to hypospadias. To date, accurate and objective evaluation of PC is not easily reproducible amongst surgeons and there are no stablished protocols on how to measure PC in a standard way and in real-time to guide intraoperative decision making. For this reason, we want to present the results of creating a semi-automated algorithm to establish a reproducible and objective assessment of PC and propose it as a standard protocol for clinical applicability using inanimate 3-D penile models.

This project consisted in two different phases. 1. Creation of an automated algorithm to estimate penile angle based on digital images. 2 Use of the algorithm to estimate penile angle on 3-D models and estimate interrater agreement using the algorithm. The algorithm was created to initially identify the geometrical centerline of the penile model to establish an automated output for angle estimation. 3-D printed penile models with known curvature angles ranging from 10 to 90° were used to test the algorithm (total of 9 penile models. These models were curved at one hinge as opposed to an arc type model. For each inanimate model, a set of 5 pictures were obtained from a lateral view at different camera angles (0⁰, 15⁰, 30⁰, 45⁰ and 60⁰) at a standard distance of 75 cm. Angle estimation using our designed PC algorithm was performed by a total of 10 different evaluators. Inter-rater reliability analysis in using the semiautomated algorithm was performed using the inter-class correlation coefficient (ICC) with two-way mixed effect model.

If the camera angle was greater than 30°, the absolute angle mean difference was greater than 10°. Camera angle with the smallest mean difference was at 0⁰ with a mean difference of 7.83°. Agreement between raters showed greater variability towards the higher camera angles. Nonetheless, a high degree of between evaluator reliability was found between the measurements at different camera angles. Single measures ICC ranges from .873 to .946, p-values were all <.0001.

Our results help standardize PC assessment using digital images and reduce subjectivity using an algorithm for PC estimation. Optimal camera position between 0⁰ to 30⁰ vertical from the penis gives the least variable and most accurate angle estimation. Future studies using algorithms will help define predictive PC cutoff values and evaluate postoperative outcomes.