Deep learning based automated quantification of urethral plate characteristics using the plate objective scoring tool...

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• The Plate Objective Scoring Tool (POST) quantifies urethral plate (UP) characteristics for surgical technique selection.
• Manual landmark identification for POST scores can introduce variability.
• Artificial intelligence (AI) offers potential for improved objectivity in hypospadias repair.

Purpose of the Study:

• To evaluate a deep learning algorithm for optimizing UP characteristic appraisal using POST on 2D images.
• To enhance objectivity and reproducibility in UP assessment for hypospadias repair.

Main Methods:

• A deep learning model was developed using a dataset of 691 images from boys undergoing hypospadias repair.
• The model performs glans localization and detects five key POST landmarks using a CNN architecture.
• Predicted landmarks are utilized for UP characteristic assessment in distal hypospadias.

Main Results:

• The model achieved high accuracy in glans localization (mAP 99.5%) and landmark detection (sensitivity 99.1%).
• A low normalized mean error (NME) of 0.07152 and mean squared error (MSE) of 0.001 were recorded.
• The landmark detection model demonstrated a low failure rate (2.5% at 0.2 NME threshold).

Conclusions:

• AI-driven POST assessment enhances standardization of UP evaluation from hypospadias images.
• Machine learning and image recognition show promise for hypospadias scoring and surgical outcome prediction.
• Further validation with multi-center international databases is underway to confirm generalizability.