IADR Abstract Archives
Multivariate Functional Shape Data Analysis: A Clinical Application
Objectives: To present a clinical application of the Multivariate Functional Shape Data Analysis (MFSDA) statistical shape method in three-dimensional (3D) reconstructions of temporomandibular (TMJ) mandibular condyles of healthy and osteoarthritis (OA) patients.
Methods: This study was composed of a sample of 92 subjects divided into 2 groups: 46 TMJ OA and 46 control. The MFSDA model considers demographics (age, gender), clinical (pain-related), biological (proteins expressed in saliva and serum samples), and imaging variables. 3D condylar surfaces for each subject are used to identify the significance of the variables and shape changes. The 3D models are aligned and mapped in a common space using SlicerSALT for the statistical analysis with MFSDA. Pearson correlation and principal component analysis (PCA) were assessed to select the most relevant features included in the MFSDA statistical model.
Results: The following variables were included in the MFSDA: Age, range of unassisted mouth opening, pain in years, PAI-1 and TIMP1 serum proteins and MMP-7 and VE-cadherin saliva proteins and the textural feature Haralick Correlation for TMJ OA group. For the control group, age, range of unassisted mouth opening, PAI-1 and MMP-3 serum proteins, MMP-7 and BDNF saliva proteins, the textural features grey level non-uniformity, and entropy were included. Our results showed statistical significance (p ≤0.05) with the shape for MMP-7 and VE-cadherin in saliva, range of unassisted mouth opening and Haralick Correlation in TMJ OA group. For the control group, the variables correlated with the shape were: age, PAI-1, and MMP-3 in serum only.
Conclusions: We showed here a novel method to correlate shape (3D reconstructions) to non-shape (saliva and serum protein expression, imaging biomarkers) features. The clinical application to evaluate the relationships between TMJ geometry with imaging, clinical and biomolecular features for TMJ OA and control patients was possible thanks to the open-source software SlicerSALT.
Methods: This study was composed of a sample of 92 subjects divided into 2 groups: 46 TMJ OA and 46 control. The MFSDA model considers demographics (age, gender), clinical (pain-related), biological (proteins expressed in saliva and serum samples), and imaging variables. 3D condylar surfaces for each subject are used to identify the significance of the variables and shape changes. The 3D models are aligned and mapped in a common space using SlicerSALT for the statistical analysis with MFSDA. Pearson correlation and principal component analysis (PCA) were assessed to select the most relevant features included in the MFSDA statistical model.
Results: The following variables were included in the MFSDA: Age, range of unassisted mouth opening, pain in years, PAI-1 and TIMP1 serum proteins and MMP-7 and VE-cadherin saliva proteins and the textural feature Haralick Correlation for TMJ OA group. For the control group, age, range of unassisted mouth opening, PAI-1 and MMP-3 serum proteins, MMP-7 and BDNF saliva proteins, the textural features grey level non-uniformity, and entropy were included. Our results showed statistical significance (p ≤0.05) with the shape for MMP-7 and VE-cadherin in saliva, range of unassisted mouth opening and Haralick Correlation in TMJ OA group. For the control group, the variables correlated with the shape were: age, PAI-1, and MMP-3 in serum only.
Conclusions: We showed here a novel method to correlate shape (3D reconstructions) to non-shape (saliva and serum protein expression, imaging biomarkers) features. The clinical application to evaluate the relationships between TMJ geometry with imaging, clinical and biomolecular features for TMJ OA and control patients was possible thanks to the open-source software SlicerSALT.
3316490_File000000.jpg
3316490_File000001.jpg
