brainscdetector:MainPage
From NITRC Wiki
Return to the main page of BRAINS
Welcome to the Wikipage for BRAINSConstellationDetector (BCD). BCD is an automated detector designed for estimating landmark constellation in 3D digital MR brain images. The tool is also one of a series in BRAINS tool suite for a comprehensive brain image analysis. It has been developed as part of my MS thesis (download here) at the Psychiatric Iowa Neuroimaging Consortium in University of Iowa Hospitals and Clinics, under the supervision of Dr. Hans J Johnson.
Contents |
Author
- Wei Lu
Motivation
Medical imaging technologies such as MRI, CT, PET, etc. enable the use of
higher resolution 3D digital image data for research and clinical treatment. The
new technologies provide improved spatial resolution at the cost of increased data
processing time. Manual identification of anatomical landmarks is still a common
practice in many neuroimaging and other medical imaging applications. It often
takes a neuroradiologist or neurosurgeon up to a few minutes to manually label a
landmark. Manually labeling of characteristic brain structures is commonly treated
as a golden standard for landmark identification, but it is labor-intensive, subjective,
and suffers from intra-/inter-rater inconsistency.
A natural and possible resolution to the difficulties of manual labeling would
be to develop a computer-aided method to conquer the challenge. However, this also
indicates that the program has to be “intelligent” enough to “understand” the scene.
Although the computer algorithm cannot know all the related domain knowledges of
neuroanatomy, pathology, physiology, psychiatry, and radiology, etc. that a human
expert would probably apply to achieve the complex human placement of landmark
points, it has to “learn” patterns to detect different landmarks from some very limited
source of information such as image intensity distribution.
An objective of this work is to provide an automated, consistent, and efficient
method of detecting important landmarks in human brain by extracting the information
of morphometric relationships among the landmarks and the landmark intensity
information in high resolution medical imaging data. The detected landmarks could
define the Talairach space that is commonly used in stereotactic neurosurgery, and
provide vital information in assisting atlas construction, tissue classification, and surface
analysis, etc. Note the landmark constellation detection method discussed in the
thesis was originally developed and optimized for brain images, but is suitable to a
larger class of medical imaging problems.
Demonstration
A detection result for cerebellum landmarks
(a) Brain atlas built by rigid transform using 3 landmarks (anterior commissure (ac), posterior commissure (pc), and midbrain pontine junction (MPJ))
(b) Brain atlas built by thin plate spline (TPS) using 41 landmarks; landmarks defined in circled area
Central slices of brain atlas built by averaging 206 transformed images. Some regions with better quality of (b) over(a) due to the additional information of landmarks are highlighted in red circles.
(c) Brain atlas built by rigid transform using 3 landmarks
(d) Brain atlas built by thin plate spline using 41 landmarks
More accurate and automated landmarks improves the atlas result: For example, in the left inset of (d), the atlas has richer information at the 4th ventricle notch, the junction of cerebral aqueduct and the fourth ventricle, and the primary fissure of the cerebellum; a clearer 4th ventricle notch and a coronal cross-sectional view of the cerebellum is given in the middle inset of (d); the ventricular head is more clear in the right inset of the same figure than its counterpart, which reflects the benefit of consistent selection of the ventricular head landmark as the most anterior cerebrospinal fluid (CSF) in left/right ventricular nucleus at the modeling phase.
(e) Brain atlas built by rigid transform using 3 landmarks
(f) Brain atlas built by thin plate spline using 41 landmarks
In (f) image regions near the optic chiasm, basal pons, genu, and rostrum, etc. are much finer than their counterparts in (e). Image regions near eyes are more clearer in the TPS-warped atlas as the information of eye centers are using in the the construction process of the atlas, which can be seen from the middle insets of (e) and (f). The cornea of eyes are crisp in the TPS warped atlas that uses more landmarks information. Finally, in the right inset of (f), the boundary of skull and brain is clearly visible thanks to the information provided by the occipital poles. Note the image regions near ac, pc, and MPJ in both the TPS-warped atlas and the rigidly aligned one have about the same quality, which is expected, as all of the three landmarks are used in both construction process.
Usage
The usage of the software can be found at our usage page.
Use cases /Tutorial
The use cases and the tutorial of the software can be found at our tutorial page.
Manuals
The manuals of the tool can be found at here.
Acknowledgment
This research was supported by funding from grants NS050568 and NS40068 from the National Institute of Neurological Disorders and Stroke, grants MH31593, MH40856 from the National Institute of Mental Health, and grant U54EB005149 from the National Institute of Biomedical Imaging and Bioengineering.
