Home: Research Projects: Image Science and Image Quality
Image Science and Image Quality
Project Leader
Eric Clarkson, Ph.D.
Project Summary
The overall goal of Core Project III is to develop practical methods for computing objective measures of image quality rapidly and to use these methods to quantify task performance for CGRI imaging systems and those of our collaborators using realistic, dynamic object models. To promote the widespread use of the methodologies to be developed under this project, the software tools, protocols, and phantoms will be made available to our collaborators and other researchers in medical imaging on our Image Quality Website.
Current Projects
- Assessment of Image Quality
- System opimization
- Kinetic SPECT imaging
- Image analysis
Methodologies for Assessment of Image Quality at the CGRI
|
Task |
Data Set |
Observer(s) |
FOM |
ComputationalMethods |
Comments |
Status as of March 2008 |
|
SKE/BKE detection |
Raw projection data |
Ideal obs, Hotelling, CHO |
AUC, SNR2 |
Pure theory (requires only mean image) |
Easy to compute, stylized task misleading |
Routine |
|
SKE/BKE detection |
Reconstructed images |
Hotelling, CHO |
AUC, SNR2 |
Noise-propagationalgorithms |
Stylized task may be misleading |
Routine |
|
SKE detection in a lumpy background |
Simulated project data |
Ideal obs, CHO |
AUC, SNR2 |
MCMC for IO, matrix inversion for Hotelling |
MCMC is time- consuming |
Routine |
|
Detection with location uncertainty |
Simulated planar images or reconstructions. |
Scanning CHO or ideal |
ALROC |
MCMC , channelized IO |
MCMC and CIO both time consuming |
Routine |
|
Estimation of activity in ROI |
Raw projection data or reconstructions. |
Wiener, channelized Wienner, MLm MAP, posterioe mean. |
Variance, EMSE or CR bound |
Object models, precomputed template, MCMC |
ROI defined independently on anatomical image |
Routine |
Estimation of tumor volume |
Raw
projection data or reconstructions. |
Wiener, channelized Wienner, MLm MAP, posterioe mean. |
Variance,
EMSE or CR bound |
Object models, precomputed
template, MCMC |
“Volume”
difficult to define |
Routine |
|
Estimation of tumor locations |
Raw projection data or reconstructions. |
Wiener, channelized Wienner, MLm MAP, posterioe mean. |
Variance, EMSE or CR bound |
Object models, precomputed template, MCMC |
|
Routine |
|
No-gold-std. evaluation methodologies |
Reconstructedimages |
Image- analysis software |
Adjusted variance |
Maximum likelihood |
Needs to be widely disseminated |
Well validated theoretically and experimentally
|
|
MRMC statistics |
ROC data |
First-principles approach |
NA |
FDA: One-shot algorithm, BGRI: bootstrap/least squares |
Passed validation tests |
Emerging as standard; combined method developed at FDA |
|
Estimation of kinetic parameters |
Dynamic projection data or reconstructions. |
Wiener, ML, GM, posterior mean, or adhoc estimators |
Variance, EMSE or CR bound |
Global compartmental kinetic model |
|
Simulations performed, physical phantom measured.
|
|
Estimation of object statistics |
Projection data for many animals |
NA
|
NA |
ML estimation via characteristicfunctional |
Uses analytic characteristic functional |
Demonstrated in simulation |
|
Detection and estimation of signal parameters |
Projection data or reconstructions |
EROC ideal, scanning MAP, ML or Hotelling |
AEROC |
MCMC, matrix methods for Hotelling |
MCMC is time-consuming |
Simulations underway |
|
SKE
detection |
Addaptive sytem
data, raw projection data |
Adaptive
Hotelling, Adaptive CHO |
SNR |
MCMC
and matrix-inversion techniques |
Probably
practical for CHO only |
Theory developed, inital simulation studies completed |
|
Estimation |
Addaptive sytem
data |
Adaptive
Wiener or channelized Wiener estimator |
EMSE |
MCMC
and matrix inversion techniques |
Probably
practical in channelized case |
Theory developed and simulation planned |
Detection
and estimation in modified MOBY phantom |
FastSPECT II |
EROC Ideal, scanning MAP, ML, or Hotelling |
AEROC |
MCMC
and matrix inversion techniques |
Probably
need to channelize |
Planned |
Key: CHO = channelized Hotelling observer, CIO = channelized ideal observer, AUC = area under ROC curve, ALROC = area under localization ROC curve, EROC = estimation ROC, AEROC = area under EROC curve, SNR = signal-to-noise ratio.
Surrogate figures of merit
(Top two figures) The performance of the Wiener estimator in the task of estimating position and size of a signal. The Wiener estimator is unable to use the image data to produce any reasonable estimates. (Bottom two figures) Performance of the scanning linear estimator for the same tasks. The new estimator is able to estimate the parameters.
Home: Research Projects: Image Science and Image Quality
 |
Center for
Gamma-Ray Imaging
October 2008 |