Correction for Scatter and Cross-talk

clangedbivalveAI and Robotics

Oct 19, 2013 (3 years and 11 months ago)

77 views

Correction for Scatter and Cross
-
talk
Contaminations in Dual
Radionuclide
99m
Tc/
123
I Imaging
Using Artificial Neural Network

Xiaoming Zheng, PhD.

School of Clinical Sciences,

Charles Sturt University

21 July, 2004

Overview


SPECT in Epileptology: Dual radionuclide
imaging.


Problems in dual radionuclide imaging:
Scatter and cross
-
talk.


The artificial neural network.


Monte Carlo Simulation results.


Phantom experimental Results.


SPECT for Epilepsy

ictal

interictal

after 24 hrs

non
-
Sz

Find the differences

Dual Radionuclide Imaging


Two time
-
points would greatly enhance the
diagnostic quality of the SPECT epilepsy images
(2nd data point could confirm seizure onset area,
seizure onset and propagation can be studied,
“connectivity” or pathways in the brain can be
characterized)


This could be accomplished by injecting two
radiopharmaceuticals in one imaging session: Tc
-
99m labeled ECD or HMPAO, and I
-
123 labeled
IMP.

0 degrees

0 degrees

0 degrees

0 degrees

90 degrees

90 degrees

90 degrees

90 degrees

Tc
-
99m window

Tc
-
99m window

I
-
123 window

I
-
123 window

Tc99m

in patient

Tc99m+I123

in patient

Problems in Dual Radionuclide Imaging: Scatter and Cross
-
talk

Problems in Dual Radionuclide
Imaging: Scatter and Cross
-
talk

-

=

0.30

Simple Correction Method

Artificial Neural Network

Neural Network Training

Momentum Backpropagation

Activation Function

Tc Scatter Correction: Training


Correction For Tc Scatter Only


Correction For Tc Scatter:

Image Differences


Accuracy vs Input Energy
Window Number

Separating Tc99m/I
-
123:

Training and Testing




Tc99m and I
-
123 Primary Images


Cross
-
talk Correction:

(Tc99m and I
-
123)


8 Input Energy Windows


Cross
-
talk Correction:


(Tc99m and I
-
123)


2 Input Energy Windows


“Oscar” Phantom Acquisition

Acquisition Protocol: Tc
-
99m

Acquisition Protocol: I
-
123

The Neural Network

Original I
-
123 Window Images From Japan

Original Tc
-
99m Window Images From Japan

I
-
123 Window Images After ANN Correction

Tc
-
99m Window Images After ANN Correction

I
-
123 Activation

Tc
-
99m Activation

Cylindrical Striatal Phantom


I
-
123

Tc
-
99m

Tc99 Window Image:

Before Correction

Tc99 Window Image:

After Correction

I
-
123 Window Image:

Before Correction

I
-
123 Window Image:

After Correction

Conclusions


The new approach of using artificial neural
network appears to be a simple but effective
method for the crosstalk correction in dual
radionuclide imaging.


Further work is to test this approach on other dual
radionuclide combinations such as Tl/Tc in
nuclear cardiology imaging.


Monte Carlo simulations on these dual
radionuclide images are currently under
investigation.

Acknowledgements


This work was supported by a Charles Sturt
University Special Study Program and NIH
research grants.


The collaborators of this work were Prof
Mike King of University of Massachusetts,
Prof George Zubal of Yale University, and
Dr John Seibyl of the Institute for
NeuroDegenrative Disorders.