Generating Simulated DT-MRI Dataset

Diffusion tensor magnetic resonance imaging (DTMRI), also known as diffusion tensor imaging (DTI), needs post processing by adequate image analysis and visualization tools. White matter tractography using DTI is becoming a routine MR technique to study white matter properties, connectivity, and alterations of fiber integrity due to pathology. The success of the method depends on the accuracy of the tracking algorithms. The disadvantage on the evaluation of such methods is that there is no gold standard regarding the true geometry of the brain anatomy or fiber bundles reconstructed in each particular case. Considering the increasing number of reports on DTI post-processing research, it is observed that most of the methodology is applied only to real data, acquired from human or animal test subjects. This approach has the following drawbacks: (i) the true anatomy of each imaged subject is not known in detail, (ii) the image noise is of little control, (iii) cost of scanner time is usually high, (iv) the comparison of the methods and the results from different studies is difficult because the data has been acquired on various MR scanner hardware, and often scanned with different imaging protocols, and finally, (v) the test data is collected from different subjects, each having unique characteristics in gross anatomy and tissue micro-architecture. In order to reduce these problems, a synthetic DTI dataset with known geometric and signal properties has been developed. The accuracy of white matter anatomical maps obtained by DTI is still unclear due to the general inability of the diffusion tensor model describing a single voxel with multiple orientational maxima. Verification and validation of the synthetic data analysis’ aims to elucidate the white matter fiber tractography in eliminating the uncertainty areas and understanding the connectivity more clearly and reliable. A model must describe how water diffuses in the synthetic dataset. For simplicity, we have in this work considered only two very basic models. In this work two separate models with different geometric properties characterized by anisotropic Gaussian diffusion are specified. These models are sampled and their output is similar to those obtained from MR scanners are generated. The efficient calculation of the diffusion tensor is achieved from that output. It is used to generate several common measures and visualizations describing Gaussian water diffusion. The project covers the geometric model, discrimination, sampling, tensor calculation, parameter calculation, and the visualization.