University of Kentucky
Are We Cartilage Experts? - Software and Experiment to Assess the Accuracy of the Segmentation for MRI Images of Knees.
Institution
University of Kentucky
Faculty Advisor/ Mentor
Jerzy Jaromczyk; Charles Staben
Abstract
Magnetic resonance imaging (MRI) is a powerful, noninvasive method that provides medical images, for use in studying and diagnosing various illnesses. Researchers at the University of Kentucky are using MRI images to study the rates and levels of decay of knee cartilage in patients, a common problem among elderly patients. Due to complexities in imaging and software processes, such as the significant amount of noise in MRI images, it is difficult to precisely identify knee cartilage. Software has been developed by Dr. P. Hardy lab at the Davis-Mills MRI UK center to semiautomatically segment (identify the boundary of) cartilage from digitized images and determine its condition. During Summer 2003, as a part of the Bioinformatics Experience for Undergraduates program at UK, as a part of a group we participated in extending and testing the software to provide additional statistical information about the accuracy of the system and its users. The package identifies a "ground truth" segmentation for an MRI image, based on the consensus selection of expert users. The process implements the maximal expectation optimization algorithm and provides and visualizes a number of statistics regarding the accuracy. With this information, we can better measure the quality of the automatic tools and the level of training, thus substantially decreasing the time required to analyze MRI images.
Are We Cartilage Experts? - Software and Experiment to Assess the Accuracy of the Segmentation for MRI Images of Knees.
Magnetic resonance imaging (MRI) is a powerful, noninvasive method that provides medical images, for use in studying and diagnosing various illnesses. Researchers at the University of Kentucky are using MRI images to study the rates and levels of decay of knee cartilage in patients, a common problem among elderly patients. Due to complexities in imaging and software processes, such as the significant amount of noise in MRI images, it is difficult to precisely identify knee cartilage. Software has been developed by Dr. P. Hardy lab at the Davis-Mills MRI UK center to semiautomatically segment (identify the boundary of) cartilage from digitized images and determine its condition. During Summer 2003, as a part of the Bioinformatics Experience for Undergraduates program at UK, as a part of a group we participated in extending and testing the software to provide additional statistical information about the accuracy of the system and its users. The package identifies a "ground truth" segmentation for an MRI image, based on the consensus selection of expert users. The process implements the maximal expectation optimization algorithm and provides and visualizes a number of statistics regarding the accuracy. With this information, we can better measure the quality of the automatic tools and the level of training, thus substantially decreasing the time required to analyze MRI images.