Ph.D. Thesis Defense - Evaluation of Chronic Wounds by Raman Spectroscopy and Image Processing
Date: March 28, 2012
Time: 2:00 PM
Location: Bossone Research Enterprise Center, Room: 709
Advisors: Ahmet Sacan, Ph.D., Elisabeth S. Papazoglou, Ph.D.
Diabetic foot ulcer has become a major health care problem as the
prevalence of diabetes and the related complications increases
globally. Due to the underlying pathological abnormalities in diabetic
patients, these ulcers usually do not heal in a timely and orderly
fashion as acute wounds do. Objective and accurate assessment of wound
healing status is needed to deliver better wound care to patients.
In this research, we utilize near-infrared Raman spectroscopy to study
tissue samples from diabetic foot ulcers on a small cohort of
patients. In our method, we categorized wounds as healing or
non-healing, harvested samples from wound debridement and collected
Raman spectra from cryosectioned samples. The average spectra from
these two groups show similar spectral profiles but the average
spectrum of healing wounds shows relatively higher intensities at
bands associated with collagen. Significant spectral features such as
individual band intensities and pairwise intensity ratios were
identified by performing unpaired t-tests between these two groups.
Supervised classification using a support vector machine (SVM)
classifier was conducted to classify the spectra or samples based on
the spectral features. The trained SVM classifier is able to predict a
spectrum’s category with 83.1% accuracy. The prediction of whether a
sample is from a healing wound can be as accurate as 95.7% or even
higher when the average spectrum of the sample was fed to the SVM
Since the quantification of the wound area is a common clinical
practice, we also applied image processing techniques to accurately
detect the wound boundary in digital images of the wound. Our method
derives from a combination of color based image analysis algorithms,
and the method is validated by comparing the performance with manually
traced boundaries of wounds in animal models and human wounds of
diverse patients. Images were taken by an inexpensive digital camera
under variable lighting conditions. Approximately 100 patient images
and 50 animal images were analyzed and high overlap was achieved
between manual tracings and calculated wound areas by our method in
both groups. The simplicity of our method combined with its robustness
suggests that it can be a valuable tool in clinical wound evaluations.
The Bossone Research Enterprise Center is located at the corner of 32nd and Market Streets.