This subproject is one of many research subprojects utilizing theresources provided by a Center grant funded by NIH/NCRR. The subproject andinvestigator (PI) may have received primary funding from another NIH source,and thus could be represented in other CRISP entries. The institution listed isfor the Center, which is not necessarily the institution for the investigator.This project aims at validating a computational, feature-learning-based framework and associated software tools for predicting central nervous system disease (CNS) using individual high-resolution MR images. We have achieved promising classification results on Alzheimers disease (AD), Mild cognitive impairment (MCI) and normal controls from a single MR image of an individual. The system will be validated for early diagnosis of AD and MCI using three invaluable datasets: a cross-sectional dataset from ADRC of UPMC, a longitudinal data set from CHS study and a Pittsburgh compound PET-MRI paired data set (PIB). The financial and social burdens of AD are compounded by recent and continued increases in the average life span. Mild cognitive impairment (MCI), emerged as a concept of prodromal AD, shows growing evidence that treatments may be most effective for improving cognitive functioning and potentially reversing pathologic changes in those individuals with MCI. Longitudinal studies indicate that persons with MCI progress to dementia at a rate of approximately 10 to 15% per year, compared with a rate of 1 to 2% per year in control subjects. In general, approximately 50% of individuals with MCI progress to develop AD over a 5-year period. This suggests that patients with MCI can follow different clinical trajectories. Thus, more sensitive and specific tools that predict future cognitive decline would strengthen accuracy in detecting pre-clinical AD and facilitate more effective early intervention. Although there are existing reported group morphological differences and rates of atrophy in human brain anatomy from serial imaging, early diagnosis of AD or MCI based on individual MRI at one time instance has challenged even the most experienced neuroradiologists. We have developed a novel approach combining computer vision, machine learning and statistical multivariate analysis into one highly automated, non-biased discriminative feature subspace discovery method that searches through the whole 3D brain. Our system is based on deformation field and tensor field features from a well-validated, state of the art deformable registration algorithm. Our preliminary studies show a leave-ten-out cross validation result on AD and control classification of 95% sensitivity and specificity. As a collaborative, interdisciplinary team of UPMC and the robotics institute of CMU, we have confidence that further validation of our current approach using a larger, longitudinal image set will make a broad impact not only in computational science and engineering but, more importantly, in translational MR image analysis for clinical practice. Currently, we run our experiments on a cluster using MATLAB code. We would like to explore developing a portable parallel code in anticipation for three larger MR image sets proposed.
Showing the most recent 10 out of 292 publications