The following is a summary of “Deep learning based automated quantification of urethral plate characteristics using the plate objective scoring tool (POST),” published in the August 2023 issue of the Pediatric Urology by Abbas et al.
The plate objective scoring tool (POST) was recently introduced as a reproducible and precise method for quantifying urethral plate (UP) characteristics and providing guidance in selecting specific surgical techniques. However, establishing the necessary anatomical reference points for calculating the POST score from captured medical images can potentially result in variability. Although artificial intelligence (AI) is yet to be fully embraced and investigated in hypospadiology, it has undeniably unveiled new potentialities. To examine the potential of utilizing deep learning algorithms to enhance the efficiency and effectiveness of evaluating urethral plate (UP) characteristics on two-dimensional (2D) medical images using the postoperative assessment to improve the objectivity and reproducibility of UP evaluation in hypospadias surgery.
The specialists accurately identified and marked the five essential landmarks in a dataset of 691 images of prepubertal boys who were undergoing primary hypospadias repair. The dataset was subsequently utilized to develop and validate a deep learning-based landmark detection model in the medical field. The suggested framework initiates with glans localization and detection, wherein the input image is cropped utilizing the anticipated bounding box. Subsequently, a profound convolutional neural network (CNN) framework is employed to forecast the coordinates of the five POST landmarks. These expected anatomical reference points are later utilized to evaluate urethral plate characteristics in cases of distal hypospadias. The suggested model effectively identified the glans region, exhibiting a mean average precision (mAP) of 99.5% and an overall sensitivity of 99.1%. A normalized mean error (NME) of 0.07152 was attained in the prognostication of the coordinates of the anatomical landmarks, accompanied by a mean squared error (MSE) of 0.001 and a 2.5% incidence of failure at a threshold of 0.2 NME.
Researchers’ findings provide evidence for the potential of establishing a standardized evaluation of UP (urethral plate) from recorded hypospadias images. Additionally, the application of machine learning algorithms and image recognition demonstrates these innovative artificial intelligence technologies’ utility in scoring hypospadias. External validation can offer valuable insights regarding the generalizability and reliability of deep learning algorithms, thereby assisting in evaluations, clinical decision-making, and prognostications for surgical outcomes. This medical deep learning application demonstrates resilience and exceptional accuracy in utilizing the POST method to evaluate UP characteristics. Ongoing evaluation using international multi-center image-based databases is being conducted for further assessment.
Source: sciencedirect.com/science/article/abs/pii/S1477513123001201
