Research Profile

imageregistration
Example of Image Registration in Medical Imaging (Whole-Body CT)
Image Registration is one of the oldest computer vision problems, that aims to find a mapping or a transformation from one image to the other. For example,  there are two images I-A (exhale respiratory motion phase) and I-B (inhale respiratory motion phase) and one can have a forward model such that one image is the same as the transformed image of the other. Then, we can solve an inverse problem to find this mapping or transformation. However, this problem is ill-posed and non-linear / non-convex. Therefore, it may yield nasty warps as shown in I-C. My past work was to design a regularizer based on diffeomorphism to alleviate this ill-posedness in the inverse problem so that more realistic warps as shown in I-D. For more information, please see “A simple regularizer for B-spline nonrigid image registration that encourages local invertibility,” IEEE Journal of Selected Topics in Signal Processing, 3(1):159-69, Feb. 2009.

My current related projects are as follows:

  • Multi-modal image registration between histology and MRI in the image domain (supported by PNUYH-UNIST Joint Research Grant for Biomedical Convergence, Collaboration with Prof Jaehyeok Lee at Pusan National University Yangsan Hospital (PNUYH) and Prof HyungJoon Cho at UNIST)
  • Multi-modal image registration between SPECT and CT in the projection domain (supported by NRF Young Investigator Research Grant)
  • Digital volume correlation (DVC) for calculating material strain and detecting cracks in high-resolution micro CT or in synchrotron imaging (Collaboration with Prof Wooseok Ji at UNIST)

 

011514_0853_MotionCorre1.png
Example of motion corrected image reconstruction in simultaneous PET-MR
Image Reconstruction is an essential step to create useful medical images from raw data in many medical imaging systems (e.g. CT, MRI, PET, SPECT).  This is also another inverse problem to estimate an original image from the raw data (e.g. projection data, Fourier samples), which is ill-posed. My past work was to study motion-compensated image reconstruction models theoretically, to correct for respiratory motion in image reconstruction in PET-MR (please see “MRI-based nonrigid motion correction in simultaneous PET/MRI,” Journal of Nuclear Medicine, 53(8):1284-1291, Aug. 2012 for details), and to design various regularizers and algorithms for better image quality. My recent research interests are to use anatomical side information (e.g. CT, MRI) to improve image quality of functional images (e.g. PET, SPECT) in filtering and image reconstruction frameworks and to use joint estimation frameworks for better image quality with limited resources.

My current related projects are as follows:

  • Improving image quality of PET or SPECT using side information such as MRI or CT in SPECT-CT and PET-MR (supported by NRF Young Investigator Research Grant, Collaborations with Profs Jeff Fessler and Yuni Dewaraja at the University of Michigan, and with Prof Jae Sung Lee at Seoul National University Hospital)
  • Joint spectral image reconstruction in Y-90 SPECT (Collaboration with  Profs Jeff Fessler and Yuni Dewaraja at the University of Michigan)
  • Dictionary learning based image reconstruction (supported by NRF Young Investigator Research Grant)

 

Example of statistical detection and various wearable biometric sensors
Statistical Detection is to use the power of statistical signal processing theories to perform various detection tasks. One example can be to detect neurons from histology images to count the number of them. Another example can be to process your ECG (heart signal) or EMG (muscle signal) to identify you. In here, statistical signal processing can play an important role to provide powerful tools to perform these tasks reliably. My past work was to use ECoG data to detect brain signals using two-covariance models (please see “Electrocorticogram as a brain computer interface signal source,” Towards Brain-Computer Interfacing, MIT Press, Cambridge, pp. 129-46, Sep. 2007).  My current research interest is to develop pattern recognition algorithms based on statistical signal processing for detecting various biological features (e.g. neurons) or for recognizing you from your unique biometric signals (e.g. ECG, EMG).

My current related projects are as follows:

  • Accurate cell counting algorithm in histology images (supported by PNUYH-UNIST Joint Research Grant for Biomedical Convergence, Collaboration with Prof Jaehyeok Lee at Pusan National University Yangsan Hospital (PNUYH) and Prof HyungJoon Cho at UNIST)
  • Pattern recognition algorithms for multi-modal biometrics (ECG, EMG) to be used as your unique password (supported by Ministry of Science, ICT, and Future Planning – IT, Broadcast Research Development Project (Main-PI: KETI), Collaboration with Profs Sung-Phil Kim (PI), Ian Oakley at UNIST)
  • Bio-Hash in statistical detection frameworks (supported by Ministry of Science, ICT, and Future Planning – IT, Broadcast Research Development Project (Main-PI: KETI), Collaboration with Profs Sung-Phil Kim (PI), Ian Oakley at UNIST)
Advertisements

Create a free website or blog at WordPress.com.

Up ↑