Automatic cystocele severity grading in transperineal ultrasound by random forest regression. (March 2017)

• Record Type: Journal Article
• Authors: Ni, Dong; Ji, Xing; Wu, Min; Wang, Wenlei; Deng, Xiaoshuang; Hu, Zhongyi; Wang, Tianfu; Shen, Dinggang; Cheng, Jie-Zhi; Wang, Huifang
• Is Part Of: Pattern recognition. Volume 63(2017:Mar.)
• Page Start: 551
• Page End: 560
• Publication Date: 2017-03
• Subjects: Cystocele grading -- Symphysis pubis detection -- Bladder boundary segmentation -- Auto-context -- Regression forest; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.09.033
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Cystocele is a woman disease that bladder herniates into vagina. Women with cystocele may have problem in urinating and higher risk of bladder infection. The treatment of cystocele highly depends on the severity. The cystocele severity is usually evaluated with the manual transperineal ultrasound measurement for the maximal distance between the bladder and the lower tip of symphysis pubis in the Valsalva maneuver. To improve the efficiency of the measurement, we propose a fully automatic scheme that can measure the distance between the two anatomic structures in each ultrasound image. The whole measurement scheme is realized with a two-phase random forest regression to infer the locations of the two structures in the images for the support of distance measurement. The experimental results suggest automatic distance measurements and the final grading by our random forest regression method are comparable to the measurements and grading scores from three medical doctors, and thus corroborate the efficacy of our method. Highlights: First automatically computerized cystocele grading method on the transperineal ultrasound is developed in this study. The automatic cystocele grading on the transperineal ultrasound is realized with a two-phase random forest regression model. Auto-context features are helpful for our regression model to improve the cystocele grading results.