Automatic Grading Of Bladder Prolapse Based On Regression Model In Pelvic Floor Ultrasound

Cystocele is a common disease in women.Accurate assessment of cystocele severity is very important for treatment options.The pelvic floor ultrasound(PFUS)has recently emerged as an alternative tool for cystocele grading.The cystocele severity is usually evaluated with the manual measurement of the maximal descent of the bladder(MDB)relative to the symphysis(SP)during Valsalva maneuver.However,this process is time-consuming and operator-dependent.In this paper,we propose an automatic regression-based scheme for cystocele grading from PFUS video.The computerized analysis of cystocele serverity can potentially improve the efficiency in the clinical practice and alleviate the workload of medical doctors for saving the time in processing and reading of ultrasound images.In this paper,we propose two regression-based methods for cystocele grading from PFUS video.First,we propose a multi-phase regression model(MPRM)to automatically evaluate the severity of cystocele.Second,we propose a spatio-temporal regression model(STRM)to improve our results of cystocele grading.It will be shown that computerized grading results are comparable to the three sets of manual rating from three medical doctors with various experiences in pelvic floor ultrasound.In MPRM,we cast the identification of the middle axis and lower tip of SP and BL as a regression learning problem from manual drawings.In the first phase,the random forest technique is applied to seek the regressional mapping from the image and geometric features to the distance feature maps of the three structures.To further consider the contextual relationship among the three structures,the auto-context method is employed to augment the feature space for the second phase training of random forest regressor.In STRM,both appearance and context features are extracted in the spatio-temporal domain to help the anatomy detection.Thus,the use of across-time-point context features can effectively impose temporal consistency on the displacement field.The evaluation of our results on 85 PFUS videos show that the MPRM-based method achieved 78.82% accuracy of the cystocele grading at best and STRM-based method achieved 87.06% accuracy of the grading at best.Meanwhile,we compared the annotations on all testing images from three radiologists to illustrate the inter-observer variation.The results show that the measurements between computer results and manual definitions are quite close to the inter-observer variation.