Master's Thesis Defense - An Automated Wound Identification System Based on Image Segmentation and Artificial Neural Networks
Date: June 13, 2012
Time: 9:00 AM
Location: Bossone Research Enterprise Center, Room: 709
Speaker(s):
Bo Song
Advisor: Ahmet Sacan
Details:
Chronic wounds are a global ongoing health challenge that afflicts a large number of people. Effective diagnosis and treatment of the wound rely largely on precise identification and measurement of the wounded tissue; however, in current clinical process, wound evaluation is based on visual inspection and manual measurements which is a subjective and inaccurate procedure. An automatic computer-based system that can perform fast and accurate segmentation and identification is desirable into the conventional and busy clinical practice, while also improving health outcomes in chronic wound care and management.
As presented in this thesis, we design such a system that uses color wound photos taken from the patients, and is capable of automatic image segmentation and wound region identification. Several commonly used segmentation methods are utilized to obtain a collection of candidate wound areas. Parameters of each method are fine-tuned by an optimization procedure. Two different types of Artificial Neural Networks (ANNs), the Multi-Layer Perceptron (MLP) and the radial basis function (RBF), with parameters decided by a cross-validation approach, are then applied with supervised learning in the prediction procedure, and their results are compared. Satisfactory prediction by the system suggests a promising tool to assist in the field of clinical wound evaluation.
Biosketch:
Directions:
The Bossone Research Enterprise Center is located at the corner of 32nd and Market Streets.
|
