Section: New Results

Image Computing: Detection, Segmentation, Registration and Analysis

Diffusion tensor imaging (DTI) is unable to represent the diffusion signal arising from multiple crossing fascicles and freely diffusing water molecules. Generative models of the diffusion signal, such as multi-fascicle models, overcome this limitation by providing a parametric representation for the signal contribution of each population of water molecules. These models are of great interest in population studies to characterize and compare the brain microstructural properties. Central to population studies is the construction of an atlas and the registration of all subjects to it. However, the appropriate definition of registration and atlasing methods for multi-fascicle models have proven challenging. This paper proposes [24] a mathematical framework to register and analyze multi-fascicle models. Specifically, we define novel operators to achieve interpolation, smoothing and averaging of multi-fascicle models. We also define a novel similarity metric to spatially align multi-fascicle models. Our framework enables simultaneous comparisons of different microstructural properties that are confounded in conventional DTI. The framework is validated on multi-fascicle models from 24 healthy subjects and 38 patients with tuberous sclerosis complex, 10 of whom have autism. We demonstrate the use of the multi-fascicle models registration and analysis framework in a population study of autism spectrum disorder. This work was performed in close collaboration with the Children's Hospital Boston.

In recent years, there have been many Multiple Sclerosis studies using longitudinal MR images to study and characterize the MS lesion patterns. The intensity of similar anatomical tissues in MR images is often different because of the variability of the acquisition process and different scanners. We proposed [29] a novel methodology for a longitudinal lesion analysis based on intensity standardization to minimize the inter-scan intensity difference. The intensity normalization maps parameters obtained using a robust Gaussian Mixture Model (GMM) estimation not affected by the presence of MS lesions. Experimental results demonstrated that our technique accurately performs the task of intensity standardization. We show consequently how the same technique can improve the results of longitudinal MS lesion detection.

Interest in quantitative MRI and relaxometry imaging is rapidly increasing to enable the discovery of new MRI disease imaging biomarkers. While DESPOT1 is a robust method for rapid whole-brain voxelwise mapping of the longitudinal relaxation time (T1), the approach is inherently sensitive to inaccuracies in the transmitted flip angles, defined by the B1 inhomogeneity field, which become more severe at high field strengths (e.g., 3T). We have proposed [26] a new approach for simultaneously mapping the B1 field, M0 (proton density), T1 and T2 relaxation times based on regular fast T1 and T2 relaxometry sequences. The new method is based on the intrinsic correlation between the T1 and T2 relaxometry sequences to jointly estimate all maps. It requires no additional sequence for the B1 correction. We evaluated our proposed algorithm on simulated and in-vivo data at 3T, demonstrating its improved accuracy with respect to regular separate estimation methods.

Quantitative MRI (qMRI) templates of relaxation times and proton density can be of particular interest for dedicated clinical applications such as characterizing brain tissue abnormalities, as well as general research purposes. We have developed in [27] 3D qMRI statistical templates consisting of T1, T2, T2* and ρ* maps from the human brain at 3T. The qMRI templates were built from a population of 20 normal controls, for which individual maps were estimated in a robust manner, accounting for acquisition artifacts and expected relationships between the relaxometry parameters. For validation, we fed the qMRI templates into a realistic MRI simulator to synthesize MR-weighted images, and compared these images with the real MR acquisitions. High correlation coefficients (>0.80) show that the developed qMRI templates can be used as input dataset for MRI simulation community, which may be of great interest to clinical neuroscience field.