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.