(Poster 168) Landmark-Based Parametric Model to Predict Pelvic Floor Muscle Geometries

Objectives: Pelvic floor muscles (PFMs) are integral to female pelvic floor function. PFM shape is related to muscle contractile and load-bearing ability and is an important indicator of injury/dysfunction. The study of PFM geometry requires segmenting individual structures from images to produce virtual 3D models. However, manual segmentation is labor intensive and time-consuming. To overcome these limitations, a parametric model was developed as an alternative approach. Ideally, this model would require only a select number of anatomic landmarks to generate a complete 3D PFM geometry.

Methods: Pelvic MRIs from nulliparous women were used to develop the model. PFMs (levator ani and coccygeus) were manually segmented for model development and later validation. The computational model was initially developed iteratively via qualitative assessment. Landmark inputs were added until there were enough coordinates to accommodate the expected PFM morphologic variability. This determined the minimum quantity of landmarks for the model to generate acceptable 3D levator ani and coccygeus muscle geometries that could then be analyzed quantitatively (Fig.1). To test the model’s accuracy, generated geometries were automatically aligned and compared to corresponding manual segmentations using surface-to-surface distances (STSDs). Although few points were used as inputs, thousands of surface points were generated to ensure representative STSDs for robust comparisons. STSDs were categorized as < 2mm, 2-5mm, >5mm. A colormap was applied to the model-generated PFMs for visualizing regions of greatest error (Fig.2).

Results: This study developed a 3D parametric model of PFMs from 13 anatomic landmark coordinates (Fig.1) and applied and compared it to 5 patient-specific manual PFM segmentations. Minimum STSDs were < 0.002mm, medians between 1.39-1.73mm, and maximums between 8.73-11.72mm. The average percentage of STSDs < 2mm were 61.33%, 2-5mm were 34.54%, >5mm were on average 4.13%. Based on STSD colormaps, the regions with least error were along the dorsal surface of the iliococcygeus and the entheses at the pubic rami. The regions with the greatest error were along the superior edges of the coccygeus muscles and inferior portion of the levator ani closest to the perineal body.

Conclusions: This study reveals that parametric models can serve as effective alternatives to manual PFM segmentation, quickly generating representative patient-specific geometries. This is an important step towards streamlining analyses of PFM morphology as this approach is computationally inexpensive and the finalized model will be easily accessible to researchers regardless of computer literacy and modeling experience. Iterative model improvement will continue by developing regions of the parametric model found to be least accurate in the present study.

Figure 1. Anterior (top) and posterior (bottom) views of a manual segmentation (left) and its corresponding model-generated geometry (right).
Figure 2. STSD colormap from model-generated geometry to corresponding manual segmentation. Anterior (top) and posterior (bottom) views of large (black) and small (gray) STSD regions.