EXPLORATION OF PARAMETER SPACE OF TWO ACTIVE CONTOUR ALGORITHMS IN THE SEGMENTATION OF MINI-PIG BURN WOUNDS
Public Deposited- Last Modified
- February 26, 2019
- Creator
-
Anand, Keerthi
- Affiliation: School of Medicine, UNC/NCSU Joint Department of Biomedical Engineering
- Abstract
- Traditional methods of burn wound analysis rely on subjective classification of shape and variable estimation of size using the rule-of-nines. For partial thickness burns, where the standard method leads to differing prognoses, it was hypothesized that the use of active contour algorithms could aid in automatic segmentation and quantitative analysis. Previous work was conducted, comparing burn wounds taken with a DSLR to a Microsoft Kinect V2 and quantifying an algorithm’s ability to distinguish the region of a burn wound on a pig. Image analysis utilizing the Edge and Chan-Vese Active Contour algorithms found the Dice-coefficient, a measure of how well a computer-generated trace aligned with a predetermined human trace, to be artificially high. It was suspected that the inflation was due to a favorable initial mask being provided to the analysis software. It was hypothesized that a difference would be noted in segmentation behavior of the two algorithms and that providing a less favorable initial trace would not yield high Dice coefficients. In order to examine the effects of starting masks on segmentation outputs and measure the generalized performance of these algorithms, in this study, we incrementally modified parameters such as the contraction bias, iteration, and starting mask, to find a parameter space that would yield repeatable, high accuracy predictions of burn wound regions. Results showed a significant difference in the performance of the algorithms and choice of starting masks at thresholds for Dice’s coefficients above 0.8 and that choice of starting mask shape and size bias segmentation. By developing an automated system to determine the best parameter set not subject to artificially high Dice’s coefficients, computers can aid physicians in accurate diagnosis and treatment of partial thickness burn wounds.
- Date of publication
- April 2018
- Keyword
- DOI
- Resource type
- Rights statement
- In Copyright
- Note
- Funding: Tom and Elizabeth Long Excellence Fund for Honors
- Advisor
- Hubbard, Devin
- Degree
- Bachelor of Science
- Academic concentration
- Biomedical Engineering
- Honors level
- Honors
- Degree granting institution
- University of North Carolina at Chapel Hill
- Graduation year
- 2018
- Language
- English
Relations
- Parents:
This work has no parents.
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
AnandKeerthi_SeniorHonorsThesis.pdf | 2019-05-18 | Public | Download |