Study design, materials and methods
This was a cross-sectional pilot study, approved by the Institutional Review Board. Radiological and functional data were obtained for female patients who underwent VUDS as part of their clinical care. Patients with active urologic malignancy, previous augmentation cystoplasty, bladder substitution, or active urinary tract infection were excluded. All studies were performed in the seated position. Demographic data were manually extracted. Single VUDS images representing a moderately full bladder (approximately 300 ml) were selected for each subject. These images were read by two independent clinicians blinded to subjects’ medical history and assessed as “normal” or “abnormal” appearing based on the presence or absence of abnormal features such as trabeculations or diverticula.
The same images were also subjected to statistical shape analysis by a blinded researcher. Shape analysis has been previously utilized in other clinical areas to determine alterations in specific features of anatomic structures. Here we applied these methods to analyze bladder shape. Bladder images were traced manually in ImageJTM and converted from an image to a 2D shape format in MathematicaTM. Corresponding points were established using DeformetricaTM, ensuring respective points on each shape represent the same anatomical landmarks. The Procrustes method was used to align the shapes as closely as possible, removing the effects of orientation, position, or global scale variation. Then a principal component analysis was performed in MathematicaTM to quantify the modes of variation, which describe shape variance and are each defined by an eigenvector and eigenvalue. In this study, modes represent shape variance in different anatomic regions of the bladder. Principal component scores were calculated for each shape and mode for use in subsequent statistical analyses.
Independent samples Student’s t-tests with Benjamini-Hochberg corrections for multiple comparisons were used to determine if clinically evaluated shape via radiographic imaging (normal vs abnormal), post void residual volume (normal vs elevated), or voiding ability (able vs unable) differed across any significant modes of variation of interest. Pearson’s correlations were performed on continuous versions of these variables on only the normally shaped bladders to evaluate differences between subgroups of interest.
Results
22 subjects have been included in this pilot study. The average age was 51 years. The primary indications for video urodynamics was severe lower urinary tract symptoms in the presence of neurogenic lower urinary tract dysfunction or previous surgery. On urodynamic testing 29% had post void residuals >300 ml. Radiographic bladder imaging was “normal” for 62.5% of subjects. All subjects had urodynamically diagnosed abnormalities of storage or emptying and severe lower urinary tract symptoms.
Shape analysis of the bladders identified 4 significant modes, which account for 43%, 19%, 14%, and 6% of the total variance in bladder shape. These modes were associated with anatomic features on radiographic imaging (Figure 1A). Qualitatively, mode 1 described changes in the proportional left to right vs superior to inferior length accounting for alterations in the lateral walls of the bladder (pink). Mode 2 and mode 3 described local variations in superior-inferior dimensions of the right (white) and left (green) walls of the bladder respectively, and mode 4 described global bladder smoothness/roundness (yellow) (Figure 1A and 1B).
Qualitative assessment of bladder modes showed separation of clinically normal and abnormal bladders for modes 1 and 3, but not for modes 2 and 4 (Figure 1B). The t-tests revealed that mode 3, which corresponds to the left superior-inferior bladder walls, differed significantly (p=0.004) for patients with clinically identified normal (-85.5±129.9) vs abnormal (149.7±210.1) bladder shapes (Figure 1C). Mode 1 was not significantly different in patients with radiographically normal and abnormal bladder shapes (Figure 1C) and there were no significant differences in post void residual volume or voiding ability for any of the groups. We then performed a subgroup analysis of subjects with radiographically normal appearing bladders; in this group a significant inverse correlation was found between mode 1 and post void residual volume with a correlation coefficient of -0.561 (p=0.046) (Figure 1D).
Interpretation of results
The results of this preliminary study indicate that shape analyses can be used to determine abnormalities of bladder shape similar to clinical assessment of radiographic images. In our study, mode 3 values representing the superior-inferior left aspects of the bladder wall reflect the clinical radiographic assessments. As clinical assessments identified trabeculations and diverticula, this mode is most associated with these features.
Secondly, our study demonstrates the ability of statistical shape analysis to detect alterations in bladder function in radiographically normal appearing bladders. Mode 1, representing the lateral bladder walls, is significantly correlated with post void residual volume in clinically normal appearing bladders. These results indicate that: 1) bladder function can be correlated to shape and 2) shape analysis is able to detect subtle variations in clinically relevant bladder shape features.
Additional VUDS data is needed to optimize the statistical shape model and begin correlation of additional shape features with bladder characteristics and function. This method demonstrates potential for the identification of more subtle alterations in bladder shape that may be associated with early diagnosis of abnormal bladder function.
This preliminary study is limited by the number or participants and that all subjects had severe lower urinary tract symptoms and urodynamically demonstrated abnormalities of either storage, emptying or both—consistent with abnormal bladder function.