The morphologic and mechanical characteristics of a tissue are fundamental to understanding the development, homeostasis, and pathology of the human body. During the previous period of funding, we developed statistical shape modeling (SSM) methods and applied these to the study of structural hip disease. We also developed the initial framework to integrate SSM with finite element (FE) analysis to enable the study of shape and mechanics together. If incorporated into clinical practice, SSM and FE analysis could identify features of the anatomy likely responsible for injury, remodeling, or repair. Geometry needed for SSM and FE models is typically generated by segmentation of volumetric imaging data. This step can be painstakingly slow, error prone, and cost prohibitive, which hampers clinical application of these computational techniques. We have created a deep machine learning algorithm ?DeepSSM? that uses a convolutional neural network to establish the correspondence model directly from unsegmented images.
In Aim 1 we will apply DepSSM to improve clinical understanding of structural hip disease by characterizing differences in anatomy between symptomatic and asymptomatic individuals; these morphometric comparisons will identify anatomic features most telling of disease, thereby guiding improvements in diagnosis. Computational advancements have simplified the process to generate patient-specific FE models, enabling clinically focused research. However, there is no framework to collectively visualize, compare, and interpret (i.e., post-process) results from multiple FE models. Currently, inter-subject comparisons require oversimplifications such as averaging results over subjectively defined regions.
In Aim 2 we will develop new post-processing methods to collectively visualize, interpret and statistically analyze FE results across multiple subjects and study groups. We will map FE results to synthetic anatomies representing statistically meaningful distributions using the correspondence model. Statistical parametric mapping will be applied to preserve anatomic detail through statistical testing. We will use our published FE models of hip joint mechanics as the test system. Finally, volumetric images provide a wealth of information that is delivered to physicians in a familiar format. Yet, tools are not available to interpret model data with clinical findings from volumetric images.
In Aim 3, we will develop methods that evaluate relationships between shape, mechanics, and clinical findings gleaned from imaging through integrated statistical tests and semi-automatic medical image annotation tools that utilize standard ontologies. Quantitative CT and MRI images of the hip, which estimate bone density and cartilage ultrastructure, respectively, will be evaluated as test datasets. To impart broad impact, we will disseminate our methods to the community as open source software that will call core functionality provided by existing, open source software that has a large user base (FEBio, ShapeWorks).
The proposed technology will provide the methodologies necessary to increase the clinical acceptance and applicability of computer models. These models measure three-dimensional tissue shape and estimate tissue mechanics, providing information that cannot be measured conventionally. We will implement these methods into software that can be used by the public free-of-charge.
Showing the most recent 10 out of 11 publications
