547 – Statistical Analysis of MRI/fMRI Data
Classification Techniques for Osteoporosis Detection from Analyzing Texture Features on Clinical Brain MRI Data
Elena G. Randou (Rantou)
FDA
Anuraag Ravikumar
George Mason University
Vasiliki N. Ikonomidou
George Mason University
Osteoporosis is a diffuse skeletal disease that is characterized by bone mass reduction and changes in the microarchitecture of the bone which eventually result to fractures, pain and disability. Magnetic Resonance Imaging (MRI) techniques, can offer insight into the fat content of the marrow and pore structure, which has been associated with osteoporosis. While these techniques are promising, the cost of MRI exam makes them an unsustainable choice for screening. We address the question of whether regular clinical brain MRI exams, can be used to identify a population at risk of osteoporosis, and allow the physician to refer them for further screening. The data set includes an osteoporosis diagnosed and a control group. Important features (texture analysis characteristics) are identified by using robust randomization tests for the difference of two means. The ability of these features for detecting osteoporosis is investigated by using different classifiers. The use of different statistical criteria provides the means of selecting the best classifier according to its performance.