Data‐driven motion estimation in Single Photon Emission Computed Tomography (SPECT)

Md. Nahid Hossain, BangladeshPh.D student, Doctoral CRP E2.40.19 

Introduction

• Single Photon Emission Computed Tomography (SPECT) is oneof the major imaging modality in nuclear medicine used fordiagnostic imaging of different organs of human body assessingphysiological function.

• A short lived radiopharmaceutical is injected into the patientbody and a series of two dimensional (2D) images ofradionuclide distribution are collected using the rotatinggamma camera.

• These data are then reconstructed for three dimensional (3D)distribution of radioactivity to give a stack of transverse slices ofthe organ or region of interest.

Cont..

Patient movement during the acquisition time is a wellknown cause of artifacts in reconstructed SPECT data.

The data acquisition occurs over a relatively long time,typically in the range of 5‐30 minutes.

The patient’s movements cause misalignment of theprojection frames, which degrades the reconstructed imageand may introduce artifacts.

These motion artifacts may significantly affect thediagnostic accuracy.

Cont..

Motion correction is the task of obtaining consistentprojection data from the acquisition.

When motion occurs between discrete acquiredprojections, consistency is lost and errors are generated inthe reconstructed estimate.

Therefore, the motion correction of patients intomography images is very much essential for accuratediagnosis and hence achieve the quality of the images .

Aims of this project Motion correction requires knowledge of the six degree of freedom motion of the head during the SPECT scan.

Previous workers have used motion tracking systems to obtain this data. 

The aim of this project is to develop and evaluate a data driven approach to motion estimation that would allow motion correction to be applied without necessity for motion tracking system.  

Simulation of motion‐affected SPECT data

Different data without motion were acquired by usinga Trionix Triad triple head SPECT camera and aHoffman 3D brain phantom filled with Tc99m inwater.

These acquired data were motion free but thephantom position was not same in all data. All datawere acquired with a little bit different position androtation of phantom to each other.

Cont.. An algorithm was developed for simulation ofmotion‐affected scans from two or three realmotion free SPECT scans by combiningprojections.

Using the simulation algorithm we can produce any desired motion‐affected SPECT scan from the acquired motion free data.

Motion‐free scans of Hoffman 3D brain phantom

Scan 1

Scan 2

SPECT data with Hoffman 3D brain phantom

Simulated data created from two different data

Simulation from another data

Simulated data (frame 51‐100)

Data 1

Data 2

Data 1

Motion corrected reconstruction

Fulton R et.al. IEEE Trans Nucl Sci 46(3); 667‐672, 1999 

First we created some simulated/test data. We createdan arbitrary using algorithm transformation.

We apply 6 degree of freedom parameters like (0,0,0)for translations and (15,8,0) for rotations to the dataand create another data.

Test data

Apply 6 dof parameter; (0,0,0) for translations and (15,8,0) for rotations and create another data

Test data with forward projection

Created moved data with forward projection

There are two SPECT scans (simulated) with the phantom in two orientations.  We could combine projections from both to  create a simulated SPECT scan with a movement at the mid‐scan point. 

We performed motion corrected reconstruction of the simulated SPECT scan in which 16 projections were acquired prior to the motion, and 16 were acquired after the motion. 

The motion will be treated as instantaneously occurring between projections 16 and 17. (Any other timing of the motion could be simulated in the way).

The next step is to perform motion correction. First we reconstruct from the first 16 pairs, to obtain a partial reconstruction

Using partial reconstruction from the first 16 pairs 

This contains the transformed partial reconstruction. We use that as the initial estimate for the next reconstruction step

Finally corrected image

Motion free image

Uncorrected image

Results The simulated data was produced by adding partialprojection data with the aim to produce a new data setso that it simulates motion induced data.

We calculated the MSD from the simulated data byanalyzing each projection frames. These showed thesignificant MSD values between two differentprojection frames which were obtained from twodifferent SPECT acquisitions.

By using the simulated data, motion correction couldbe performed.

Mean Square Difference (MSD) Calculation

Mean Square Difference (MSD) Calculation

Next steps To trial a recently algorithm for data driven motion 

estimation in CT  on SPECT phantom data with known motion. 

Optimize the algorithm for SPECT data.  Apply and evaluate this algorithm on clinical DAT‐SPECT 

data acquired for dopamine transporter imaging.  Perform an observer study to assess the clinical impact of 

motion correction on diagnosis.  Identify any limitations of the method e.g. sensitivity to 

noise, inability to correct for motion during projection. 

Thank you all