Predicting 2D Geometrical Shapes Using Partial Contour Data and Principal Component Analysis
Open Access
- Author:
- Ziegler, Alexandria
- Area of Honors:
- Mechanical Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Daniel Humberto Cortes Correales, Thesis Supervisor
Daniel Humberto Cortes Correales, Thesis Honors Advisor
Jason Zachary Moore, Faculty Reader - Keywords:
- Statistical Shape Model
Principal Component Analysis
Partial Data Predictions
Ultrasound
Spinal Anatomy
Cross-correlation - Abstract:
- The inaccuracy of spinal needle placement drastically diminishes the treatment of patients with chronic, debilitating back pain. With that, it is imperative to formulate a method to create a more accurate representation of spinal anatomy from limited ultrasound data. The primary goal of this study is to determine if a statistical shape model is an accurate predictor for the shape of an unknown image. Utilizing airplanes, numerous images were inputted to create the shape model which includes an average image associated with numerous principal component modes. From there, a separate image not included in the training set was inputted to a prediction function utilizing the average image and mode shapes from the statistical shape model. To understand the accuracy of the model with partial data, the airplane outline was manipulated to have data removed. From the study, it was found that about 1/3 of the outline points can be removed while still preserving an accurate prediction. For vertical slicing, up to 65% of the image could be removed while still preserving a highly accurate prediction. As for top-down horizontal slicing, the prediction was accurate through up to 70% of the airplane being removed. The bottom-up slicing was much less accurate, being accurate only through 35% of the airplane being removed. This is likely due to the heavy dependence on the shape of the tail within the prediction algorithm. The findings of this study support the extension into the 3D prediction of spinal anatomy from ultrasound imaging data. With the extension, it is important to identify key landmark features on spinal anatomy, as well as to include a significant number of images in the training set to ensure high accuracy of the predictions.