Artificial intelligence models derived from 2D transperineal ultrasound images in the clinical diagnosis of stress urinary incontinence

Introduction and hypothesis The aim of the study was to develop artificial intelligence (AI) algorithms using 2D transperineal ultrasound (TPUS) static images to simplify the clinical process of diagnosing stress urinary incontinence (SUI) in practice. Methods The study involved 400 patients in total, including 265 SUI patients and 135 non-SUI patients who underwent a routine clinical evaluation process by urologists and TPUS. They were classified into different groups based on the International Consultation on Incontinence Questionnaire (ICIQ) to assess the impact of inconvenience on patients’ lives. Four AI models were developed by 2D TPUS images: Model A (a single-mode model based on Valsalva maneuver images to classify G-0, G-1, and G-2); Model B (a dual-mode model based on Valsalva maneuver and resting state images to classify G-0, G-1, and G-2); Model C (a single-mode model based on Valsalva maneuver images to classify G-2 and G-01); Model D (a dual-mode model based on Valsalva maneuver and resting state images to classify G-2 and G-01). The performance of the four models was evaluated by confusion matrices and the area under the receiver-operating characteristic curve (AUC). Results The dual-mode model based on the Valsalva maneuver and resting-state images ( Model D ) had a higher accuracy of 86.3% and an AUC of 0.922, which was significantly higher than the AUCs of the other three models: 0.771, 0.862, and 0.827. Conclusions The AI algorithm using 2D TPUS static images of the Valsalva maneuver and resting state may be a promising tool in the diagnosis of SUI patients in to relieve clinical processes in practice given its ease of use in clinical applications.