Home    中文  
 
  • Search
  • lucene Search
  • Citation
  • Fig/Tab
  • Adv Search
Just Accepted  |  Current Issue  |  Archive  |  Featured Articles  |  Most Read  |  Most Download  |  Most Cited

Chinese Journal of Endourology(Electronic Edition) ›› 2024, Vol. 18 ›› Issue (06): 597-605. doi: 10.3877/cma.j.issn.1674-3253.2024.06.011

• Clinical Research • Previous Articles     Next Articles

Application of artificial intelligence model based on pelvic floor ultrasound for the diagnosis of female stress urinary incontinence

Junlong Huang1, Wenshuang Li1, Xiaoyang Li1, bolong Liu1, Yilong Chen2, Huiping Qiu2, Xiangfu Zhou1,()   

  1. 1.Department of Urology,the Third Affliated Hospital of Sun Yat-sen University,Guangzhou 510630,China
    2.Guangzhou Tianpeng Computer Technology Co.,Ltd 510000,China
  • Received:2024-10-18 Online:2024-12-01 Published:2024-11-26
  • Contact: Xiangfu Zhou

Abstract:

Objective

Based on two-dimensional transperineal pelvic floor ultrasound images obtained during the Valsalva maneuver,this study aims to develop a deep learning artificial intelligence model to assist in the clinical grading diagnosis of female stress urinary incontinence (SUI).

Methods

All patients underwent transperineal pelvic floor ultrasound examination within one month of their initial visit. According to the International Consultation on Incontinence Questionnaire-Short Form (ICI-QSF),the SUI patients were divided into three groups: mild (S1),moderate (S2),and severe (S3). Baseline information and ultrasound images of the patients were collected,and midsagittal plane images of the pelvic floor during the Valsalva maneuver were selected. After preprocessing,the regions of interest(ROI) were semi-automatically segmented using computer algorithms. A transfer learning approach was applied using a ResNet50 model pretrained on ImageNet. The data were randomly divided into a training set and a test set in an 8∶2 ratio,with five-fold cross-validation performed on the training set. After model training,performance was evaluated using a confusion matrix and the area under the receiver operating characteristic curve (AUC). Finally,gradient-weighted class activation mapping (Grad-CAM) was used to generate heatmaps to visually demonstrate the areas of interest identified by the model.

Results

A total of 282 patients were included in the study,comprising 167 patients with SUI and 115 non-SUI patients(S0). The model achieved an accuracy of 84.1% in the training set and 83.9% in the test set. The AUC values in the training set were as follows: non-SUI 0.89 (95%CI: 0.78-0.97),mild SUI 0.96 (95%CI: 0.90-1.00),moderate SUI 0.92 (95%CI: 0.81-1.00),and severe SUI 0.93 (95%CI: 0.84-0.98). In the test set,the AUC values were non-SUI 0.88 (95%CI: 0.77-0.98),mild SUI 0.91 (995%CI: .80-0.99),moderate SUI 0.92 (95%CI: 0.80-1.00),and severe SUI 0.88 (95%CI: 0.75-0.98).

Conclusion

The grading diagnostic artificial intelligence model based on two-dimensional transperineal pelvic floor ultrasound images during the Valsalva maneuver shows potential to become an objective basis for the grading diagnosis of SUI patients.

Key words: Stress urinary incontinence, Pelvic floor ultrasound, Convolutional neural network, Deep learning, Artificial intelligence

京ICP 备07035254号-20
Copyright © Chinese Journal of Endourology(Electronic Edition), All Rights Reserved.
Tel: 020-85252990 E-mail: chinendourology@126.com
Powered by Beijing Magtech Co. Ltd