|
|
|
|
|
RESEARCH |
| |
| EXPLOSIVE DETECTION | ||
 |
For stand-off (i.e. without contact) explosive detection at airports and postal offices, it is assumed that a person preparing or carrying explosive substances will inadvertently contaminate him/herself or the exterior of the package. To detect such traces of explosive materials we are using a differential reflectometry (DR) device. The DR device provides hyper-spectral data for each pixel that is measured, and in this hyper-spectral data, explosives show characteristic behaviors at specific wavelengths. For example, the signal strength of TNT shows a dramatic drop at 420nm. However, as the amount of the explosive decreases and as the distance of the probed material to the sensor increases, the signal-to-noise ratio decreases, and it becomes harder to observe the characteristic behaviors. To overcome these challenges and to detect trace amounts of explosives, I am developing a system that combines traditional classifiers and end-member detection techniques for sub-pixel detection. This project is in collaboration with the MSE department at UF. WHISPERS'2011 GHC'2011 | |
 | ||
| | ||
| LANDMINE DETECTION | ||
|
Over thirty-nine countries - mostly spread into Africa, Asia, Europe, Middle East and South America - suffer from the threat of currently buried 60 million landmines and around 26000 people a year are wounded or killed by these mines. In this study, we are interested in the detection of landmines using robotic systems with ground penetrating radar (GPR) and wide-band electromagnetic induction (WEMI) sensors. The WEMI sensors are good at finding the mines with metallic content; whereas the GPR sensors are good at finding the plastic mines. However, landmine detection is still a hard problem because of the various sizes and different types of mines; as well as the fact that their signature changes with temperature, humidity and soil conditions. In this study, we are interested in all aspects of the problem from noise removal to feature extraction, to context based learning, decision and feature level fusion, and classification. IGARSS'2008 ICPR'2010 PRRS'2010 |
|
| | ||
| | ||||
| DIFFUSION TENSOR MRI VALIDATION USING FLUORO DATA | ||||
 |  |  |
||
| Most of the previous efforts on enhancing the DT-MRI estimation/smoothing have been based on "what is assumed to be correct"; and there are only very few studies concentrating on the validation of these approaches. This project presents
our current efforts and observations for a validation framework. In the scope of this framework, high resolution fluoroscopy slices obtained from the brain stem of a rat are compared with the fibers tract maps obtained from the processing of the Diffusion Tensor Magnetic Resonance images of the same rat. Several steps prior to comparison involve (i) the segmentation of the fibers from the fluoro data and
the registration of the fibers for 3D stacking; (ii) processing of the DT-MRI data to obtain the fiber tracts; (iii) 3D registration of the fluoro volume to the DT-MRI data.
REPORT'2006 |
||||
| | ||||
| | ||
| TRACE INFERENCE | ||
 |
Upon our previous efforts to validate the DT-MRI fiber tracks with fibers extracted from fluoroscopy data, we fell into a problem with fiber crossings. Basic DT-MRI fiber tracking methods fail at the crossings and need to be inferred from the data. In this study, the trace is estimated from tangents and curvatures, with a number of constraints defined in a neighborhood.
REPORT'2007 |
|
| | ||
| DETECTION OF ACUTE REJECTION FROM DYNAMIC CONTRAST ENHANCED MRI | ||
 |
 |
 |
| Acute rejection is the most important cause of renal dysfunction after kidney transplantation. As the rejection develops, transplanted kidneys develop abnormal flow patterns that are not uniformly distributed throughout the whole kidney. Quantification of these changes via image analysis can be used to detect rejections, thus replacing risky biopsy procedures. This study investigates the potentials of Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI) as a noninvasive imaging technique to differentiate acute rejection transplants from normal functioning transplants; and focuses on finding image analysis techniques for the registration and segmentation of renal images. The outcome of this research is an image analysis framework embedded into a graphical user interface to be easily tested by the doctors. While additional follow-up is necessary, the software is ready in pure C++ language with a user friendly GUI (using Qt). The whole project is compilable under Windows, Unix and Linux. MICCAI'2006 ISBI'2006 CARS'2005 MASTER THESIS'2005 |
||
| | ||
| SHAPE BASED SEGMENTATION USING DEFORMABLE MODELS | ||
 |
Surgical planning, navigation, medical visualization and diagnostics all benefit from image segmentation. However, segmentation is still a challenge because of the image noise and inhomogeneities; therefore, segmentation algorithms can not depend only on image information but also have to exploit the prior knowledge of shapes and other properties of the structures to be segmented. In this study, a novel 3-D level sets segmentation approach is proposed that uses both the gray level and shape information. CVPRW'2006 MICCAI'2005 |
|
| | ||
| EARLY DETECTION AND FOLLOW-UP OF LUNG NODULES | ||
 |
Our long term research goal is to develop an automatic approach for early detection of lung nodules that may lead to lung cancer. Approaches for the early detection of lung cancer consist of two main steps: 1) Early detection of lung nodules from either low dose spiral computed tomography (LDCT) or x-ray, and 2) Tracking the behaviour of detected lung nodules over time. This paper focuses on the monitoring of the progress (growth or shrinking) of lung nodules in successive chest low dose CT (LDCT) scans of a person using non-rigid registration.
CARS'2005 |
|
| | ||
| INTERACTIVE 3D GRAPHICS | ||
 |
This was a very cool project to model my face. I got the triangulation from a 3D scanner in Dr. Jorg Peters's Lab, which also gave the 2D texture image of my 3D scan. Then I placed a control mesh for a NURBS Surface, and mapped the texture image on the surface. By playing with the control points of a spline, I got the mesh to fit my 3D scan.
|
|
| | ||
|
|