Improved Image Processing Algorithm for Non-Contact Indentation of Artificial Skin Samples
Ty Parker, Jason Yao, PhD, and Stephanie M. George, PhD
Department of Engineering, East Carolina University
- Our team is developing AERO to improve lymphedema assessment.
- AERO delivers a consistent air impulse against the skin, captures the indentation with a camera, and processes the images to quantify tissue response.
- The previous image processing algorithm required manual adjustment and had difficulty under certain lighting effects and with sample variation.
- Optimize AERO’s image processing algorithm to handle low-light conditions and variable human factors.
- Design a more effective and automatic image processing algorithm.
New Image Processing Algorithm
Future Directions and Conclusions
- In comparison to the old process, the new algorithm uses increased contrast, 2D convolution, and a series of safety checks to better identify and smooth the indentation border, as well as removing unwanted objects and shadows from the frame.
- The new algorithm offers improved automation.
- As early identification is important, the overlap between levels 1 and 2 will be investigated.
- More testing is needed to conclusively find the optimal data collection conditions.
- Next steps include testing the algorithm on a larger variety of samples with diverse skin color, angle, and lighting conditions to find conclusively which method produces the best identification of the indentation.
- The long-term significance of this work is increased early detection and intervention of lymphedema.
- Funding from the National Science Foundation, EEC-1659796 – REU Site: Biomedical Engineering in Simulations, Imaging and Modeling (BME-SIM); Undergraduate Research and Creative Achievement Award (Dingeman); Interdisciplinary Research Award (Yao); NSF I-Corps Funding; and Summer Ventures.
- Keith Williams, Antonia Dingeman, Grace Krell, Joshua Dupaty, Harrison Gillespie, Sonya Hardin, James Buck
- For both algorithms, the maximum area and area of the curve (AUC) increase with increasing edema level (Table 1).
- The maximum area results from the new algorithm and old are statistically, significantly different for all 4 edema levels; while for AUC on for levels 1 and 4 (p<0.05).
- For the new algorithm, maximum area was statistically different between edema levels (p<0.05).
- Note the new algorithm increased the difference between level 3 and 4 but also had more overlap between levels 1 and 2.
Table 1. The table above shows the difference in maximum area and area under the curve between the old and new algorithm for each edma level.
Figure 2. The figure shows the image as it is processed at several different stages.
- Data was collected from four Lifeform artificial phantoms with varrying levels of edema.
- Experimental set-up: ambient light; nozzle angle of 60 degrees and pressure of 80 PSI.
- Two operators collected five trials per sample for a total of 40 trials.
1. Videos are broken down into frames
2. The images are converted into grayscale and their contrast is enhanced in order to better identify the darkest region, which corresponds to the indentation
3. These enhanced images are then binarized and 2D convolution is used to blur and subsequently re-threshold the image, smoothing the indentation’s border and eliminating small, noise-like objects
4. A series of safety checks then follows to make sure all objects have been cleared from the image except for the indentation to the indentation
5. Note, the blue line representing the indentation boundary. Indentation data such as the start and end time, decay, total area, and area under curve are recorded.
Figure 3. Maximum area by edema level for the new algorithm results (left) to the old algorithm results (right).
Figure 1: a) AERO set-up; B) captured indentation; and C) binarized image using old algorithm, Note the rough boundary.