A Study On Image Quality Improvements For MRI-Guided Diffuse Optical Tomography

As its advantages of non-invasiveness, portability, and easiness to use, Diffuse Optical Tomography (DOT) has great potential in brain functional examination for special crowds, such as patients in intensive care unit (ICU), children and infants. However, it still has a long way to go in clinical applications for its low spatial resolution caused by the ill-conditioning and ill-posed problem in the DOT reconstruction procedure. It is significant to improve the imaging performance for its widespread acceptance in brain function diagnosis.In this thesis, we utilize the head anatomical structural information from magnetic resonance imaging (MRI) to build a Finite Element Model and use the model to guide the DOT reconstruction procedure. This approach can improve the low spatial resolution and image localization errors resulted by the hemispherical head model or the general head template, ill-conditioning, and ill-posed problems in the reconstruction. The procedures of modeling the human brain are as follows: firstly, in view of the features of MR image, we combine multiple segmentation methods, such as thresholding, hybrid level set and mathematical morphology, to classify the whole head into five parts, scalp, skull, cerebrospinal fluid, gray matter and white matter; Secondly, We implement robust algorithms for meshing and labeling segmented images according to different tissues. A volumetric mesh generation toolkit is developed for medical images; thirdly, we register the spatial coordinates of MR and DOT imaging system respectively using a Least Square method. Finally, with structured and unstructured volume rendering algorithms, the three-dimensional visualization and interaction are provided for both MRI and DOT images.Simulations and in vivo experiments are completed to evaluate the image quality of MRI-Guided Diffuse Optical Tomography. Using point spread function simulation, we provide a thorough evaluation of the image quality varying the optical parameters, classification strategies and the numbers of the mesh node. It is indicated that a four-layer brain model (scalp, skull, brain and cerebrospinal fluid) produces the best image quality. With the optimal tissue classification strategy and phase-encoded visual stimuli, we have obtained the cortical activations from DOT system which are identity with physiological significance.