The Cyclops Mean Brain Map Project was started in 2002 by Daniel Duarte Abdala as a Master Thesis research subject and actually is in its second research cycle.This web page presents a brief overview of the research activities I have performed during my M.Sc. studies at the Cyclops Project and the results that were achieved at their conclusion. It also presents the actual research developments.
Project Overview
With the improvement of computational power as well as with the new methodologies for image alignment, it has become possible to interpret a huge amount of imaging data from the human brain.
It would be a very interesting possibility in the assessment of degenerative cerebral diseases to have access to a “mean brain”, a statistically representative description of brain tissue characteristics of a given population, which as well could be used as a basis for clinical decision making.
This work presents the results of two and a half years of research and development as a member of The Cyclops Project with the objective to create a methodology for the volumetric image registration and consequent mean brain map generation. As a subproduct of this work, there was developed a methodology for the cortical atrophy evaluation, which can be used both in a global and local fashion. This work was performed with the intention to aid neuroradiologists on the process of early Alzheimer’s disease diagnosis, as well as in assessment and prediction of the statistically representative MCI to Alzheimer evolution.
Objectives
The main objective of this work was to develop and start to validate a methodology to perform the creation of statistical templates of human brain tissue density to address and represent the common characteristics of a given population, aiming to aid the early diagnosis of Alzheimer’s disease as well as to allow the chronological evolution follow up of the cognitive loss in patients with MCI – Mild Cognitive Impairment – syndrome.
To perform this research, it was necessary to develop and adapt several mathematical and computational approaches and techniques.
Specific Objectives
1. Analyze the existing methods related to image registration;
2. Analyze the mathematical models to template reference values definition;
3. Develop a cerebral normalization methodology – aiming at the transformation of a given image/voxel volume to a stereotaxical standard space;
4. To develop a methodology for the mean brain generation based on normalized volumes;
5. To develop a methodology to analysis the regional and global atrophy grad of the cerebral cortex;
6. To Model a computational system that allows one to create generic templates for a given population, based on the previously proposed methodology;
7. To Model a system for the evaluation of previously generated mean brain maps to verify the regional/global atrophy of Alzheimer complainer patients.
Material & Methods
The methodology for the development of mean brain models and the framework for the evaluation of regional/global atrophy grades had as a study background a initial population supplied by University Hospital Johannes Gutenberg, in Mainz, Germany. The data set was compounded by 61 subjects (24 male, 37 female) showing ages between 51 and 77 years.
All patients were previously anonymized using the Cyclops DCM Anonymizer, and were examined using a Siemens Magneton Sonata with 1.5 Tesla and using a , high resolution T1 MPRage series with TR=9.7s, TE=4s, a fov. of 22cm. Results were represented as matrices of 256×256 pixels x 180 slices in Sagittal orientation , with isometric voxels of 1.0 mm³.
From among the 61 subjects, a set of 26 patients who presented a healthy diagnosis confirmed by neuropsychological tests and who showed no inclination to further dementia development (11 male and 15 female) was used for the mean brain creation.
The remaining 35 patients presented a diagnosis as follows:
• Seven among then, (3 male, 4 female) was classified as MCI subjects;
• Two subjects, (1 male, 1 female) was classified as Vascular Dementia subjects;
• Ten subjects, (2 male, 8 female) was classified as Dementia Complainers;
• Sixteen subjects, men (3 male, 13 female) was classified as Alzheimer subjects;
These patients were used to create atrophy maps and to perform measurements for the global/regional cortex atrophy during the methodology evaluation process.
We also had produced atrophy maps and measured the cortex global/regional of the control subjects to calibrate and validate de quality of the produced mean map.
Work Description
The main idea behind this work is to produce a mean map of a given population etymologically selected by age, actual diagnosis state and predisposition to development of any kind of neurodegenerative disease.
There are a number of applications in neuroscience where a comparison between a previous exam and an actual one is desirable and can possibly lead to better diagnosis. By example, in Multiple Sclerosis, chronological comparisons between exams could lead to a better understanding of how the disease is evolving, with functional brain areas are being more affected by disease evolution as well can answer questions how a given medicine is acting by visual inspection of it’s differences. The same is true to Dementia, in special to Alzheimer Dementia, where the loss of gray matter is statistically significant but there are no standard of what’s the non disease state since there are no previous exam of a given patient before the disease’s appearance. Image 1 presents a schematic example of chronological performing comparison.
With no previous exams it’s impossible to perform such comparisons, and to fill this lack of information, a mean brain that statistically represent a given population, can be used in place of a previous exam. To it works, it’s necessary to restrict the subjects used to produce the mean brain, and also is necessary to restrict such comparisons to only patients that lay within population characteristics. Image 2 presents a mean brain usage to such comparisons.
However, to perform such comparison is necessary to realign or register the exams. To perform it, there are a number of methodologies described in literature.
In a mono-subject, mono-modality application, where the main source of misalignment is positional and possible changing in gray matter, the registration task becomes mathematically trivial. Image 3 and Image 4 presents respectively a mono-subject, mono-modality example before and after alignment.
However, to create mean maps, we will be dealing with a multi-subject, mono modality problem. To solve it, the chosen procedure was to perform the realignment in two steps:
• Normalization to a standard stereotaxic space – To perform this, we executed a set of 3D geometrical transformations to match the volumes. The solution was to find a set of anatomical points and match them using NICP – Normalized Iterative Closest Point [2], a variation of the standard ICP algorithm;
• Registration – For the correction of possible misalignments that still persisted, registration based on a context based approach was performed.
The control points identification follows a novel approach developed by us, where intra and inter cerebral anatomical points were selected, especifically the base of the fourth ventricle and the center of the eyeballs. At a fisrt glance, this seems to be a good assumption, but makes the method impractical to be applied in subjects with eye’s deformation.
The method also follows a search space strategy that takes advantage of the gold proportion (1,618033…) in which the control points always lay within the red, green and blue regions presented in Image 5.
After all subjects have been registered, there was applied a voxelwise computation to estimate the mean and standard deviation values. With the mean values there was created a mean brain volume and the standard deviation values were the basis to create color maps using the color scale presented in Image 6.
Within the content of this work, the color maps represent an estimative of which voxels have signal value that deviates from the expected value on the mean map. Our thesis is that the deviations below the mean expected value can be used to infer which regions are experiencing some level of atrophy. Image 7 and Image 8 present an example of color map results. They show respectively a health patient and an Alzheimer Dementia patient from the University Hospital of Mainz.
