This dissertation presents a new computational system that seeks to estimate the inverse mechanics of a bone fracture from 3D CT images. One image provides a model of an unbroken bone and a second records a fractured example of the same bone after an injury. For a bone fracture, these mechanics specify how the bone broke in terms of what object impacted the limb and how the bone fracture fragments moved over time due to that impact. To accomplish this task, novel image processing techniques are used in combination with existing computational dynamics simulation tools. Estimates of a fracture event are generated in three steps: (1) estimate a 3D model of the limb immediately before the fracture event occurred, (2) iteratively simulate the fracture dynamics to search for values of the unknown fracture variables that produce fracture patterns similar to that depicted in the observed image of the fractured limb, and (3) visualize and analyze likely fracture scenarios in a virtual 3D environment. This dissertation investigates the following two challenges: (1) obtaining a physically-meaningful estimate of the unbroken limb and (2) solving the difficult search problem for the unknown fracture event variables. Challenge one requires geometric surface models having unprecedented accuracy to be estimated for the fracture fragments which is a difficult unsolved problem. Challenge two requires conceptualizing and implementing a completely new system to estimate the fracture event. Results for the proposed methods are provided for clinical tibial plafond fracture data.
