PI: Jianhui Zhong, PhD. Co-Investigators: Xing Qiu, PhD and Jeffrey Bazarian, MD
Brain is consisted of highly complex network of neurons and pathways. Impact of brain injury often involves multiple brain areas that are highly variable both in locations and time evolution in individual patients. A longitudinal analysis of neuroimages in individual patients is therefore mandatory for the diagnosis and treatment. We are using diffusion tensor imaging (DTI) to scan a group of patients who are admitted into Emergency Room at the Strong Hospital with mild traumatic brain injury (mTBI) within 24 hrs of accident and subsequently three times at one week, one month, and three months after the injury. Specific natures of such study give rise to two new challenges and consequently require a new paradigm for image analysis:
(1) A need for quantitative analysis of DTI images and DTI-derived parameters of individuals acquired longitudinally. Neuroimages such as DTI typically consist of tens of thousands of image voxels (individual brain spatial samples) yet only a few measurements of such brain volumes are typically acquired. Due to inter-dependence of data within each brain volume yet very limited number of independent data points, the conventional statistical analyses are not readily available. It is also important to eliminate unavoidable variations in scanners, and normal physiological variations in individual at different time of measurements, so that true pathological changes reflected by the longitudinally acquired images can be quantified.
(2) A need for quantification of intrinsic connectivity and changes of it due to brain injury within each individual brain. Tractography of interconnected brain regions via axonal bundles can be derived from DTI data to describe these characteristics. Generation of such tractograph, however is very often prohibited by limited signal-to-noise ratio in data, and is computationally very demanding.
We propose to address the above challenges with the following two Specific Aims:
Aim 1: A modified wild-bootstrap analysis (WBT) will be used to simulate multiple repeated measurements, for the individual patient data at each time point, to derive measurement variations.
Aim 2: A mixed ANOVA model will be used to separate contributions of signal changes from different sources (scanner stability, changes in normal individual physiology, and true changes due to pathology).