Real-Time Typical Urodynamic Signal Recognition System Using Deep Learning
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Purpose: Gold-standard urodynamic examination is widely used in the diagnosis and treatment of lower urinary tract dys function. The purpose of urodynamic quality control is to standardize urodynamic examination and ensure its clinical refer ence value. In our study, we attempted to use a deep learning (DL) algorithm model, mainly for the recognition of typical uro dynamic signal, to help physicians complete high-quality urodynamic examinations. Methods: Urodynamic image data from 2 cohorts of adult patients with neurogenic bladder were used: (1) 300 patients with neurogenic bladder in our center from 2012 to 2018 (1,960 images used to train and validate the DL model); and (2) 100 pa tients with neurogenic bladder from 2020 to 2021 (695 images used to test the performance of the DL model). This resulted in a total of 2,655 images to train, validate and test the DL algorithm to predict the urdynamic signals. Results: Yolov5l had the best detection performance and the highest comprehensive index score (F1, 0.81; mean average pre cision, 0.83). Our study is a retrospective single-center study, and the generalization ability of the model has not been verified. Conclusions: DL algorithms can help operators identify typical urodynamic signals in real time, improve the interpretation and quality of urodynamic examination, and benefit patients.
Abstract
Purpose: Gold-standard urodynamic examination is widely used in the diagnosis and treatment of lower urinary tract dys function. The purpose of urodynamic quality control is to standardize urodynamic examination and ensure its clinical refer ence value. In our study, we attempted to use a deep learning (DL) algorithm model, mainly for the recognition of typical uro dynamic signal, to help physicians complete high-quality urodynamic examinations. Methods: Urodynamic image data from 2 cohorts of adult patients with neurogenic bladder were used: (1) 300 patients with neurogenic bladder in our center from 2012 to 2018 (1,960 images used to train and validate the DL model); and (2) 100 pa tients with neurogenic bladder from 2020 to 2021 (695 images used to test the performance of the DL model). This resulted in a total of 2,655 images to train, validate and test the DL algorithm to predict the urdynamic signals. Results: Yolov5l had the best detection performance and the highest comprehensive index score (F1, 0.81; mean average pre cision, 0.83). Our study is a retrospective single-center study, and the generalization ability of the model has not been verified. Conclusions: DL algorithms can help operators identify typical urodynamic signals in real time, improve the interpretation and quality of urodynamic examination, and benefit patients.