Research Showcase
The images above represent core research projects currently undertaken by the BMIT Research group.
Click a thumbnail above to view the details of its corresponding project.
New modelling techniques for non-invasive quantifications
Achievements: To estimate the physiological parameters, kinetic modelling requires frequent sampling of arterial blood samples (as the input function) for several hours. We pioneered input-function and cascaded modelling, which extracts physiological parameters and the required input function simultaneously from images.
Scientific novelty: Our proposed techniques indirectly extracted information embedded in the image data whereby creating a new direction in system modelling.
Impact: This avoids the need for invasive blood samples and revolutionised the field of functional imaging. Patients are no longer subject to painful blood sampling during scanning which not only help financially, but also removes the risk of exposure to radiation and possible blood infection (e.g. AIDS).
K.P. Wong, D. Feng, S.R. Meikle & M.J. Fulham, "Simultaneous Estimation of Physiological Parameters and the Input Function: In-vivo PET Data" IEEE Trans. Info, Tech. Biomed., vol. 5(1), pp. 67-76, 2001.
X. Li, D. Feng, K-P. Lin, S-C. Huang, "Estimation of myocardial glucose utilisation with PET using the left ventricular time-activity curve as a non-invasive input function", Medical & Biological Engineering & Computing, vol. 36(1), pp. 112-117, 1998.
Data compression
Achievements: Reducing the size of data from expontentially-increasing medical imaging is essential for telecommunication and real-time processing. We pioneered a data compression category, namely diagnostic lossless data compression (DLLDC), achieving a compression ratio up to 100:1, while maintaining diagnostic accuracy in functional imaging.
Scientific novelty: DLLDC was achieved by changing the fundamental optimization criteria from subjective visual perception accuracy to objective diagnostic quantification accuracy with the removal of redundant information in temporal, spatial and frequency domains.
Impact: DLLDC provides enabling technology for integrated e-health care systems by allowing vast amounts of medical image data to be stored and accumulated online (rather than on tapes and DVDs) to better facilitate data sharing and data mobility.
D. Ho, D. Feng, & K. Chen, "Dynamic image data compression in spatial and temporal domains: Theory and algorithm," IEEE Trans. Inform. Technol. Biomed., vol. 1, pp. 219-228, 1997.
D. Feng, W. Cai & R. Fulton, "Dynamic Image Data Compression in Spatial and Temporal Domains: Clinical Issues and Assessment", IEEE Trans. Inform. Technol. Biomed. vol. 6(4), pp. 262-268, 2002.
Data acquisition
Achievements: We developed the optimal image sampling schedule (OISS) theory, superior to traditional strategies based on trial and error.
Scientific novelty: The famous Shannon Sampling Theorem, which opened up a new area of digital signal processing, is used only for uniform samples of non-noisy signals. Our OISS has been a significant step forward in extending digital signal-processing, and provides the theoretical foundation for the optimal scheduling of general non-uniformly sampling of noisy-analogy signals.
Impact: Our OISS has become a standard data acquisition method for functional imaging in biomedical/pharmacological research and clinical applications. It has reduced sample size by more than 80%, while retaining all the qualitative and quantitative information of the original analogy signals.
Z. Chen, D. Feng, W. Cai, & R. Fulton, "Performance Evaluation of Functional Medical Imaging Compression via Optimal Sampling Schedule Designs and Cluster Analysis", IEEE Trans. Biomedical Engineering, vol. 52(5), pp. 943-945, 2005.
C.H. Lau, S. Eberl, D. Feng, H. Iida, P-K. Lun, W-C. Siu, Y. Tamura, G.J. Bautovich, & Y. Ono, "Optimized acquisition time and image sampling for dynamic SPECT of Tl-201", IEEE Trans. Med. Imag., vol. 17(3), pp. 334-343, 1997.
Biomedical functional parametric imaging
Achievements: Parametric imaging provides functional quantitative information at voxel-by-voxel level. However, computational complexity involved with parametric images meant that unrealistic assumptions or simplifications had to be adopted, and thus leading to biased/limited estimations. We developed the world's first generic unbiased, statistically reliable and fast technique called generalized linear least square (GLLS) algorithm.
Scientific novelty: GLLS innovatively converts complicated non-linear estimation into simple linear estimation for non-uniformly sampled, continuous biomedical system identification.
Impact: GLLS is widely used in functional images for the quantification of biochemical processes such as cerebral/myocardial blood flow, oxygen utilization rate, glucose computation rate, and distribution volume of neuro-receptor.
D. Feng, S-C. Huang, & Z. Wang, "An unbiased parametric imaging algorithm for nonuniformly sampled biomedical system parameter estimation", IEEE Trans. Med. Imag., vol. 15(4), pp. 512-518, 1996.
L. Wen, S. Eberl, D. Feng, W. Cai & J. Bai, "Fast and reliable estimation of multiple parametric images using an integrated method for dynamic SPECT", IEEE Trans. Med. Imag., vol. 26(2), pp.179-189, 2007.


