Quantitative 3D Analysis of Levator Ani Muscle Subdivisions in Nulliparous Women: MRI Feasibility Study

Usage of different euqipment

For the image acquisition all MRIs have been perfromed on either a 1.5T MRI scanner (Signa® EXCITE EchoSpeed HD or HDx. GE Healthcare, Waukesha, Wisconsin, USA; used until 12/2014) or a 3T MRI scanner (MAGNETOM Skyra. Siemens Healthcare GmbH, Erlangen, Germany; used from 01/2015). The MRI scans were screened and manually reviewed. MR-scans which did not cover the necessary landmarks or where the landmarks were not visible due to degraded image quality were excluded. The MRIs from both scanners were comparable. Although MR images at 3.0T typically have a higher in-plane resolution compared to 1.5T images, up to date there is no clear evidence that clinical MRI scans at 3.0T improve the diagnostic performance in a clinical scenario (References: 37, 38). Also our MRI scans were performed for clinical purposes and with the same slice thickness (3 mm) at 1.5 T and 3.0T, which is a crucial parameter for precisely delineating structures (as the levator ani) that run through the acquisition plane.  The detailed imaging parameters are listed in the Material and Methods section.

Nevertheless, the use of two different MRI field strength may have introduced a certain bias. We included a respective statement in the limitations section (highlighted in green).

Sample size

This is a technique development study to improve the understandig of Leavtor ani muscle in MRI and it is a first quantitative description of the muscles morphometry within the 3D PICS system. We acknowledge the relatively small cohort.
As adressed in our discussion section, we acknowledge the need for larger patient cohorts to establish normative values for the Levator ani muscle, as well as to provide precise statements about the ICC. Additionally, as mentioned in the discussion, it is evident that newer MRIs offer superior image quality, thereby contributing to more precise point placement. Such considerations should undoubtedly be taken into account in future investigations

Limits of agreement reporting

A Bland-Altman plot was examined by the authors on all aggregated data points to assess overall absolute bias, linearity of bias, and proportional bias (partially investigating a potential Simpson's paradox), but was not included in the manuscript.

The ICC statistic has the advantage of being a single number, which is meaningful to provide an informative overview, as we have many combinations (different muscle points, spatial dimensions, body sides, etc.).
In contrast, the Limits of Agreement (LoA) provides the reader with a more comprehensible scale. However, as this measure depends on the scale, it also depends on the muscle points and on the spatial dimension, it must be assessed for each spatial point for instance (the only assumption we found justifiable was that left/right is interchangeable - also checked graphically). We have therefore included a table showing the absolute (mean) bias between the two raters, the lower and upper LoA together with the 95% Wald CI.

It is also important to remember that any potential absolute bias, proportional bias, and LoA found between the two raters refers to people, not to two different types of methods/machines. This means that two different raters could have different levels of bias, or the same raters could reduce differences after feedback rounds. The aim of the agreement analysis (ICC, Bland-Altman) was to show that the agreement was generally good enough to build further studies/analyses on.

Of course, the number of paired observations is a relevant limitation for both methods. Assumptions such as homoscedasticity, absence of proportional bias, or normality cannot be properly checked.

We encourage future researchers, based on the results of this feasibility study, to consider sample sizes that will allow a more accurate assessment of the general agreement possible between raters. This pilot study can then ideally be used to estimate the sample size required for this purpose.

à We have included a Table in the Appendix section (Table A2: absolute (mean) bias between two rates, the lower and upper LoA with the 95% Wald Confidence Interval in brackets) Additionally, we have incorporated comments into the material & methods, results, and discussion sections. We kindly ask you to review the highlighted paragraphs (yellow) for any necessary revisions.

Time required for the assessment and potential clinical implementation

At the outset of the project, the primary author familiarized herself with the anatomy of the levator ani muscle. The anatomical landmarks were positioned based on anatomy-based instruction (Table A1). Initially, the marking of these points took slightly longer; however, a steep learning curve was observed over time, allowing for the efficient identification and marking of the respective points in a short period. Marking all points of interest (n=29) thus took less than 30 minutes.  Subsequently, the marked points, identified in the Osirix DICOM Viewer, were converted into the 3DPICS format.The raw data was captured and exported to Microsoft Excel. A JavaÒ tool converted this raw data to the 3D-PICS coordinate system (x’/y’/z’) inside the excel sheet. As our project evolved, the 3D-PICS program now allows the direct extraction of the 3D-coordinates relative to the PICS plane. Bony landmarks have to be set once for each MR dataset as a reference frame. This additionally simplified the process. At present the Identification of anatomic landmarks is not yet suitable for clinical implementation. We set groundwork for further research and future studies with lager sample size. The next research project undertaken by part of the the study group involves automating segmentation and creating a user-friendly application for 3D-PICS. This application aims to facilitate widespread and straightforward utilization of the 3D-PICS technology. A preoperative staging using MRI, which provides an accurate depiction of pelvic anatomy, holds promise for enhancing surgical precision and improving patient outcomes. Moreover this new knowledge will advance the study of the physiologic ranges of supportive structures and organ locations and thus help to define threshold values so as to distinguish physiological from pathological alterations before clinically relevant prolapse is obvious.