Results
As stated in the introduction, the objectives of the project were:
- Create a program to classify the 3D trackers within a CT scan.
- Create a program to classify bones within a CT scan.
- Create a program that uses the bone and tracker locations as input, and returns the linear transformation needed to relate the trackers to the bone.
Objective 1 was fully satisfied by the Tracker Detection Algorithm. The TCA uses a CT scan as an input and returns eight ℝ3 coordinates of the form: (𝑥,𝑦,𝑧). These coordinates are used as an input in a later objective.
Next, Objective 2 was satisfied by the Bone Detection Algorithm. Using the same CT scan as input, the BCA prompts the user to make selections based on the type of surgery performed. Combining the processed CT scan with the user inputs, the BCA returns two ℝ2 coordinates (assuming the XY plane was selected) of the form: (𝑥,𝑦).
Last, Objective 3 was satisfied using the outputs from Objective [1] and Objective [2] as input into the Transformation Calculator. Using the eight coordinates from the TCA, two planes are created (one for each tracker). Next, the points created by the BCA are appended, such that two additional planes are created (one for each bone of interest). Using matrix algebra, a transformation is created such that each of these sets of planes can be transformed to align with one another.
Deliverables
The goals for the deliverables included a working spine navigation algorithm that enables orthopedic surgeons to assess a patient’s misaligned bones during the corrective procedure. We were trying to meet all three objectives to reach the ultimate goal. We succeeded in creating a program to classify the 3D trackers within a CT scan, named BCA. The original plan for creating a program to classify bones within a CT scan was to use an interactive UI and ask the doctor to choose the line of interest in any direction. We shifted the plan to exploring machine learning algorithms to improve the accuracy and automatic the program to minimize user input. However, with the limitation in time, the machine learning algorithms were not ideal. Therefore, we switched back to functional BCA and planned to improve the machine learning BCA in the future. Third, we have created a program to match the trackers from camera reading to CT scan. The final report and website will still be a deliverable and complete project and leave many possible future directions and guidelines to improve and automate the design. All the code has been uploaded to GitHub for future reference.