Quantitative assessment of new MRI-based measurements to differentiate low and high stages of pelvic organ prolapse using support vector machines
- 184 Downloads
Introduction and hypothesis
The objective of this study was to quantitatively assess the ability of new MRI-based measurements to differentiate low and high stages of pelvic organ prolapse. New measurements representing pelvic structural characteristics are proposed and analyzed using support vector machines (SVM).
This retrospective study used data from 207 women with different types and stages of prolapse. Their demographic information, clinical history, and dynamic MRI data were obtained from the database. New MRI measurements were extracted and analyzed based on these reference lines: pubococcygeal line (PCL), mid-pubic line (MPL), true conjugate line (TCL), obstetric conjugate line (OCL), and diagonal conjugate line (DCL). A classification model using SVM was designed to assess the impact of the features (variables) in classifying prolapse into low or high stage.
The classification model using SVM can accurately identified anterior prolapse with very high accuracy (>0.90), and apical and posterior prolapse with good accuracy (0.80 – 0.90). Two newly proposed MRI-based features were found to be significant in the identification of anterior and posterior prolapse: the angle between TCL and MPL for anterior prolapse, and the angle between DCL and PCL for posterior prolapse. The overall accuracy of posterior prolapse identification increased from 47 % to 80 % when the newly proposed MRI-based features were taken into consideration.
The proposed MRI-based measurements are effective in differentiating low and high stages of pelvic organ prolapse, particularly for posterior prolapse.
KeywordsPelvic organ prolapse Dynamic MRI Classification
Conflicts of interest
S. Onal: None.
S. Lai-Yuen: None.
P. Bao: None.
A. Weitzenfeld: None.
D. Hogue: None.
S. Hart: Speaker and Consultant for Boston Scientific, Covidien, Cooper Surgical, and Stryker.
S Onal: manuscript writing, model building, and data analysis
S Lai-Yuen: project development, manuscript writing/editing
P Bao: project development
A Weitzenfeld: project development
D Hogue: data collection
S Hart: project development, data collection, manuscript editing
- 6.Gousse AE, Barbaric ZL, Safir MH (1998) Dynamic “HASTE” MRI sequence in the evaluation of all female pelvic pathology. Urology 159:328–334Google Scholar
- 14.Robinson CJ, Swift S, Johnson DD, Almeida JS (2008) Prediction of pelvic organ prolapse using an artificial neural network. Am J Obstet Gynecol 199:193.e1–193.e6Google Scholar