35
Comparison of Single Shot Methods for R2* Comparison Presentation for Kana Lab, Lab Meeting Rishi Deshpande Thesis Committee: Dr. Donald Twieg, Chair Dr. N. Shastry Akella Dr. Georg Deutsch University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009 1

Comparison of Single Shot Methods for R2* Comparison

  • Upload
    sissy

  • View
    59

  • Download
    0

Embed Size (px)

DESCRIPTION

Comparison of Single Shot Methods for R2* Comparison. Presentation for Kana Lab, Lab Meeting Rishi Deshpande Thesis Committee: Dr. Donald Twieg, Chair Dr. N. Shastry Akella Dr. Georg Deutsch. - PowerPoint PPT Presentation

Citation preview

Page 1: Comparison of Single Shot Methods for R2* Comparison

Comparison of Single Shot Methods for R2* Comparison

Presentation for Kana Lab,Lab Meeting

Rishi Deshpande

Thesis Committee: • Dr. Donald Twieg, Chair• Dr. N. Shastry Akella• Dr. Georg Deutsch

University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009 1

Page 2: Comparison of Single Shot Methods for R2* Comparison

Outline

Introduction Basics of MRI, fMRI Significance of reliable R2

* estimation Single-shot methods: MEPI and SS-PARSE

Experiment and Analytical Methods Trajectory generation Data acquisition Reconstruction and comparison of accuracy and temporal variability

Discussion

Conclusion

Future scope

2University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 3: Comparison of Single Shot Methods for R2* Comparison

1H nuclei within tissues 1H nuclei under external magnetic field

RF pulse (sinc/Gaussian/square)

1H get dislodged from steady state . They release energy while returning to steady

state.

Energy is collected as a function of 2D-Inverse

Fourier Transform

Sources:http://www.cs.sfu.ca/~stella/main/_spins_figure8.gifhttp://www.cs.sfu.ca/~stella/main/_spins_figure7.gifhttp://videos.howstuffworks.com/discovery-health/14537-human-atlas-mri.jpghttp://www.mr-tip.com/exam_gifs/brain_mri_transversal_t2_002.jpg

Applying a 2D-FFT to the signal data

generates 2D-images in the imaging plane.

Basics of MRI Image Acquisition

3University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3rd , 2009

Page 4: Comparison of Single Shot Methods for R2* Comparison

Control/Stimulation acquisition

Estimation of Neuronal activity↓

BOLD effect ↓

R2*

fMRI

4University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 5: Comparison of Single Shot Methods for R2* Comparison

BOLD Response Model:

Significance of reliable R2* estimation

5

*BOLD = Blood Oxygenation Level Dependent* R2

* = 1/T2*

University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 6: Comparison of Single Shot Methods for R2* Comparison

R2* Measurement: Multiple Shot Method

Gradient Echo Multiple Shot (GEMS)

Echoes can be closely stacked, thus enabling accurate R2* calculation

Serves as a gold standard in the comparison study

6University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 7: Comparison of Single Shot Methods for R2* Comparison

Single Shot Methods

Multiple Gradient Echo – Echo Planar Imaging (MEPI)

7University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 8: Comparison of Single Shot Methods for R2* Comparison

SS-PARSE

Conventional model Estimates map: M(x)

M(x) w(x) R2* (x)

SS-PARSE model

Include local phase evolution & local signal decay

Estimate maps (images) of M(x), R2* (x), ω(x) by solving an inverse problem.

It uses Progressive Length Conjugate Gradient (PLCG) algorithm which requires optimal parameters to minimize least squared residuals to generate parameter maps.

Single-Shot Parameter Assessment by Retrieval from Signal Encoding

8

Censored for gratuitous math

Censored for gratuitous math

University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 9: Comparison of Single Shot Methods for R2* Comparison

Encoding Strategy (k-trajectory)

k space k, t spaceModeling Acquired Data

Comparing Conventional MRI & SS-PARSE Methods

Decoding StrategyInverse FFTInverse Estimation

9University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 10: Comparison of Single Shot Methods for R2* Comparison

10

Source: http://commons.ucalgary.ca/at-wld/images/cartoon02.gif

University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 11: Comparison of Single Shot Methods for R2* Comparison

Project Goals - experimental Create gradient waveforms and generate trajectories for 7 different gradient

strengths (1.9 G/cm to 3.8 G/cm):

Implement the sequence on Varian 4.7 T vertical scanner using phantoms

Compare performance of SS-PARSE with MEPI based on:1. Accuracy of R2

* estimates (compare with Gradient-Echo results)2. Temporal variability of R2

* (over time-series of 50 acquisitions)3. Find correlation between R2

* and TSD values4. Find correlation between maximum gradient strength and accuracy

Gmax = 1.9 G/cm Gmax = 3.8 G/cmLower k-space coverage Larger k-space coverage

Fewer data points More data points

Faster parameter estimation Slower parameter estimation

11University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 12: Comparison of Single Shot Methods for R2* Comparison

Project goals – Theoretical Inferences

Factors contributing towards performance of SS-PARSE:

1. Gmax values – Find relationship between • Gmax and R2

* estimates (compared with gradient-echo values)

2. Shimming – Find effects of field inhomogeneity in SS-PARSE and MEPI studies.

3. Performance over R2* range - Observe the changes in temporal

behavior over R2* values typically found in human brain tissues

(20 to 40 sec-1 in 4.7 T MRI systems)

12University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 13: Comparison of Single Shot Methods for R2* Comparison

65 ms

1.9 G/cm 2.29 G/cm 2.5 G/cm 2.9 G/cm3.2 G/cm 3.5 G/cm 3.8 G/cm

k-trajectory Generation and Calibration

Calibration data acquired at: ±2, ±4, ±6, ±8, ±10, ±12 mm displacements in x & y planes

For Gmax:1.9, 2.29, 2.5, 2.9, 3.2, 3.5 and 3.8 G/cm.

13University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3rd , 2009

Censored for gratuitous math

Page 14: Comparison of Single Shot Methods for R2* Comparison

Phantom For Data Acquisition

R2* Range: 15 sec-1 to 45 sec-1

14University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3rd , 2009

Page 15: Comparison of Single Shot Methods for R2* Comparison

Data Acquisition: GEMS, MEPI and SS-PARSE

1. SS-PARSE acquisitions• Per study = (7x Gmax) x (50x repetitions)• Repetition time = 5 second• Slice Thickness = 3 mm

2. MEPI acquisitions• Per study = 50x repetitions at 4 echo times• Resolution = 64 x 64• Repetition time = 5 second• Echo Times = 22.3, 66.8, 96.4 and 124.2 millisecond• Slice Thickness = 3 mm

3. GEMS acquisitions• Per study = 16 x echo times• Resolution = 128 x 128• Echo Times = 5, 8, 10, 12, 15, 20, 25, 30, 35, 40, 45, 50,

55, 60, 65 and 70 millisecond• Slice Thickness = 3 mm

Performed total 18 experiments to obtain the R2* values in the desired range (15 to 45 sec-1)

Hardware: 4.7 T 60 cm-vertical-bore Varian primate MRI system (Varian Inc., Palo Alto, CA)

15University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 16: Comparison of Single Shot Methods for R2* Comparison

Source: http://www.hagencartoons.com/cartoon159.gif

16University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 17: Comparison of Single Shot Methods for R2* Comparison

Development of GUI For Analysis & File Handling File Handling PLCG Tweakers Parameter Maps

17University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 18: Comparison of Single Shot Methods for R2* Comparison

R2* Evaluation: GEMS and MEPI

•R2* is computed over a ROI

• •Monoexponential fitting of signal to echo times.

MEPI

GEMS

18University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 19: Comparison of Single Shot Methods for R2* Comparison

Parameters Estimates in SS-PARSE Reconstruction (SS-PARSE)

Parameter maps were computed using the PLCG algorithm from all the SS-PARSE acquisitions. Maps were created for all Gmax values (1.9 G/cm to 3.8 G/cm).

19University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 20: Comparison of Single Shot Methods for R2* Comparison

Accuracy of R2* Estimation

1. R2* estimates from SS-PARSE and MEI plotted vs. R2

* from GEMS2. Ratio of R2

* accuracy plotted vs. R2* estimates from GEMS

SS-PARSE and MEPI estimates and accuracy plot at SS-PARSE Gmax = 2.9 G/cm

20University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 21: Comparison of Single Shot Methods for R2* Comparison

Accuracy Over Gradient Amplitudes

Accuracy of R2* estimation computed by using the ratio:

|R2* MEPI - R2

* GEMS |

|R2* SSPARSE - R2

* GEMS|

was > 1 for following percentage points over the Gmax range:

1. 1.9 G/cm: 61.3%2. 2.29 G/cm: 64.2%3. 2.5 G/cm: 66.4%4. 2.9 G/cm: 68.3%5. 3.2 G/cm: 67.6%6. 3.5 G/cm: 65.6%7. 3.8 G/cm: 61.2%

Accuracy of estimation (ratio) was maximum at Gmax = 2.9 G/cm

21University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 22: Comparison of Single Shot Methods for R2* Comparison

Temporal Variation of R2* Over 50 Repetitions

TSD computed for:

• Each pixel over 50 repetitions

• Each ROI over 50 repetitions

• For MEPI and SS-PARSE

• For Gmax with best accuracy

22University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 23: Comparison of Single Shot Methods for R2* Comparison

23University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Mark with lower temporal variability,Thus lower TSDGood

Mark with higher temporal variability,Thus higher TSDNot Good

Page 24: Comparison of Single Shot Methods for R2* Comparison

Depiction of TSD in MRI Studies

24University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd, 2009

Page 25: Comparison of Single Shot Methods for R2* Comparison

TSD Plots

The difference was > 0 for 79.3% to 97.3% for R2* values between 15 sec-1 and 45 sec-1

• Dot indicates TSD at a single pixel

• Each blob of pixels represents a tube with a different R2*

• Scatter plot for the difference TSD(MEPI) – TSD (SS-PARSE) shows points around the difference = 0 line

• Dots above the difference=0 line show locations where the performance of SS-PARSE was better than of MEPI

R2* (GEMS) vs. TSD (SS-PARSE)

R2* (GEMS) vs. TSD (MEPI)

R2* (GEMS) vs. [TSD (MEPI) and TSD (SS-PARSE)]

25University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd, 2009

Page 26: Comparison of Single Shot Methods for R2* Comparison

Performance Under Field Inhomogeneity

MEPI

SS-PARSE

26University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd , 2009

Page 27: Comparison of Single Shot Methods for R2* Comparison

Parameter Estimation Under Field Inhomogeneity

SS-PARSE parameter maps have an one-on-on correspondence with the ROI from GEMS image (obtained before intention deshimming)

MEPI image appears distorted in one direction and the ROI does not correspond with ROI from GEMS. Even though we have studied the behavior of MEPI, the same behavior is also observed in standard EPI scans, which is the common modality used in clinical fMRI sudies.

R2* computation in MEPI is impossible under field-inhomogeneity because of a noticeable

geometric distortion.

27University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 28: Comparison of Single Shot Methods for R2* Comparison

Source: http://www.yachigusaryu.com/blog/pics/sci_principles_cartoon.jpg

28University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd, 2009

Page 29: Comparison of Single Shot Methods for R2* Comparison

Conclusions

Gradient waveforms for seven Gmax values were developed for SS-PARSE and were used to acquire phantom data

Performance of SS-PARSE and MEPI was compared using GEMS as the gold standard (for accuracy and TSD) over range of Gmax values.

Performance of SS-PARSE improved with increasing gradient amplitude until 2.9 G/cm. Thereafter the performance deteriorates.

SS-PARSE has a lower TSD than MEPI. This means it can estimate the parameters much reliably over several repetitions when used in fMRI studies.

SS-PARSE is able to reconstruct reliable parameter maps even in the presence of field inhomogeneities.

29University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 30: Comparison of Single Shot Methods for R2* Comparison

Future Scope

PLCG algorithm requires adjusting the algorithm tweakers heuristically. With better knowledge about the estimation process we should be able to set the parameters in a deterministic manner.

With better problem conditioning, and with MRI systems capable of delivering more than 6.5 G/cm (hardware limit of Varian 4.7 T system), we should be create trajectories with much higher sampling rates, thus giving accurate parameter estimation.

Parallel acquisition and multiple shot trajectories, increases the number of sample points, thus improving conditioning of the inverse problem and leading to more accurate estimates.

30University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009

Page 31: Comparison of Single Shot Methods for R2* Comparison

Acknowledgement

Advisor:Dr. Donald Twieg

Committee MembersDr. N. Shastry AkellaDr. Georg Deutsch

Dr. Stan Reeves (Auburn)

CDFI & VSRC colleagues:

Mark BoldingRajiv MenonNingzhi LiMatt WardDebbie WhittenJerry Millican

Parents and Sister

FriendsMichelleJonChris

Grant Support:

NIH # R21/R33 EB003292

City of Birmingham

31University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd, 2009

Page 32: Comparison of Single Shot Methods for R2* Comparison

Thank You

32

http://upload.wikimedia.org/wikipedia/commons/b/b1/Thumbs_up_by_Wakalani.jpg

University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd, 2009

Page 33: Comparison of Single Shot Methods for R2* Comparison

Questions

Source: http://www.lifehack.org/wp-content/files/2007/12/question.jpg

33University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd, 2009

Page 34: Comparison of Single Shot Methods for R2* Comparison

Extras

34University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd, 2009

Page 35: Comparison of Single Shot Methods for R2* Comparison

Cramer-Rao Lower Bound for standard deviation of error for Multiple-Echo EPI (MEPI) and Rosette, SNR=200

Rosette (k,t)-trajectories acquire more information on R2* than multiple-echo EPI trajectory

MEPI Rosette

Idealized radial

s.d.

of R

2*

R2* (sec-1)

35University of Alabama at Birmingham, Department of Psychology, Lab Meeting Presentation, September 3 rd 2009