15
Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours Dr S J Doran Department of Physics University of Surrey S 1 Department of Physics, University of Surrey, Guildford 1 C Domenig, 2 A Jurasz, 3 M Leach, 1 S Doran 2 Glaxo Smith Kline 3 Clinical MR Research Group Institute of Cancer Research

Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

  • Upload
    wilona

  • View
    38

  • Download
    0

Embed Size (px)

DESCRIPTION

S. 1 Department of Physics, University of Surrey, Guildford. 2 Glaxo Smith Kline 3 Clinical MR Research Group Institute of Cancer Research. 1 C Domenig, 2 A Jurasz, 3 M Leach, 1 S Doran. Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours. Dr S J Doran - PowerPoint PPT Presentation

Citation preview

Page 1: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Dr S J Doran

Department of PhysicsUniversity of Surrey

S 1Department of Physics,University of Surrey,Guildford

1C Domenig, 2A Jurasz, 3M Leach, 1S Doran

2Glaxo Smith Kline3Clinical MR Research GroupInstitute of Cancer Research

Page 2: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Structure of talk

• ADC as a measure of treatment response:a tantalising prospect

• Why Burst imaging for diffusion?Why not Burst imaging!

• Initial analysis of the data

• Further analysis of the data and future work

Page 3: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

A tantalising prospect: Diffusion imaging in tumours

• Intriguing measurements were made using the novel Burst diffusion imaging sequence.

• These appeared to show that (in this patient cohort) there is a very strong link between treatment outcome and ADC prior to treatment.

• However, there were a number of issues concerning the methodology that required further investigation.

• This talk is about what we found as we delved deeper into the data.

Lancet 360, 307–308 (2002)

Page 4: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

IVIM Measurements in tumours

• Previous studies have evaluated ADC’s in extra-cranial organs using only a restricted range of b-values, sometimes as few as two.

• The existence of a significant tissue perfusion effect is intrinsically of interest.

• Moreover, if the existence of perfusion is ignored, then incorrect values of the ADC may be calculated.

• Measurement with multiple b-values is relatively time-consuming and few studies characterise the low b-value regime fully.

Yamada et al., Radiology, 210, 617–623 (1999)

Results in liver

Page 5: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Why use Burst for extra-cranial diffusion imaging?

• Measurement of diffusion coefficients using Burst was first introduced in 1995.

• Burst allows us to obtain a very large number of points on the diffusion decay curve.

• This gives the potential for analysing multiple exponential signal decay.

• This form of Burst leads to images without distortion: potentially much more suitable for extra-cranial imaging than EPI.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 5 10 15 20 25 30 35 40

Echo Number

A /

A0

Data for CuSO4

T2 and D double fit

Doran and Décorps, JMR A, 117(2), 311–316 (1995)

Page 6: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Why not Burst imaging?

• Burst uses low flip angle pulses, so the SNR is very poor.

• Although typically 9-25 b-values are acquired in the same time as a single PGSE b-value, this is still a multi-shot technique.

• This gives rise to motion artifacts, as in PGSE, that may compromise our data.

• We need to compensate for T2

decay during the acquisition.

Page 7: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

• SNR was too poor to make a good quantitative analysis on single pixels.

Initial analysis of the data Anomalously high D for fat is due to T2 “correction”. Standard multi-echo sequences measure an incorrect T2 for fat.

b-value / s mm-2

ln (

S/S

0)

• However, the results for tumour ROI’s appeared very promising, leading to a good quality fit.

Tumour regression / %

AD

Cm

on

o /

cm

2 s

1

r = -0.83, p = 0.012• A “naïve” automated analysis,

based on a single exponential diffusion diffusion decay led to the results published in The Lancet.

Page 8: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Further analysis of the data (1)

• Closer examination showed that not all tumours followed the same pattern.

• A single-exponential diffusion decay model was clearly inappropriate for most.

• The data are fitted moderately well by a bi-exponential model.

• This suggested that IVIM effects may be important.

b-value / s mm-2

ln (

S/S

0)

S/S0 = f exp(-b.ADCbiexp) + (1-f) exp(-bD*)

Page 9: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Further analysis (2): Key questions

This observation poses a number of significant questions:

• What did we actually measure?

• How do we get a genuine ADC from these measurements?

• How much of what we see is due to the low SNR of Burst?

• Are the results caused by incorrect T2 measurements in our

“correction scan” or motion artifacts?

Page 10: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Further analysis (3): What did we measure?

b-value / s mm-2

ln (

S/S

0)

Effect of original analysis was to return an average between ADC and D*. Not so very different from doing a two-point diffusion measurement!

• However, results are severely biased by where the cutoff is chosen.

• Fitting a single-exponential decay to only the first half of the semi-log plot allows us to make a crude estimate of the pseudo-diffusion coefficient for individual pixels.

b-value / s mm-2

ln (

S/S

0)

• Fitting to the last half of the plot gives us an estimate of ADC.

Page 11: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Further analysis (4): SNR issues

• Ideally, we would always perform a double exponential fit.

• SNR is too poor to do this on individual pixels, but we can fit a straight line to get D* for every pixel.

• We have a wide spread of values, but how much of this is genuine and how much due to low SNR?

Conclusion 1: The effects that we see are not artefacts of low Burst SNR

D* / 103 mm2 s1

Nu

mb

er

of

pix

els 128 128

64 64

32 32• We can increase SNR by rebinning the

data to lower resolution

• With SNR increased by factors of 2 and 4, we maintain the broad range of D*.

Page 12: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Further analysis of the data (4)

• To our surprise, we found no correlation between D and D* as obtained in this model with tumour regression.

• One patient had an anomalously high value for D* and was tentatively excluded from our subsequent analysis.

Tumour regression / %

AD

Cb

iex

p /

10

3 m

m2 s

1

r = 0.03, p = 0.012

Tumour regression / %

D*

/ 1

03 m

m2 s

1

r = 0.14, p = 0.143

Conclusion 2: The (genuine) effect seen is not caused by D, as at first thought.

• We then fitted an IVIM diffusion model to data for the tumour ROI’s.

Page 13: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Further analysis of the data (5)

• We did find a correlation (albeit relatively weak) between diffusion fraction f and tumour regression.

• This correlation is consistent with the original observation that ADCmono measured with a mono-exponential model

decreases with increasing tumour regression.

r = 0.61, p = 0.012

Tumour regression / %

Dif

fus

ion

fra

cti

on

Page 14: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Discussion

• We still do not understand fully the origin of the excellent correlation in our original result.

• The parameter originally measured is a combination of ADC and perfusion.

• The “diagnostic” parameter appears to be the diffusion fraction, f, rather than ADC or D* per se.

• Further volunteer studies have highlighted the large sensitivity to motion of this un-navigated sequence.

• There are some concerns that any mis-estimation of T2 in our

data correction could mimic a multi-exponential behaviour in the data.

Conclusion 3: It is difficult to envisage how the possible systematic errors above could have led to the correlation seen.

Page 15: Intravoxel Incoherent Motion Imaging in Locally Advanced Rectal Tumours

Conclusions

• We have measured a very interesting phenomenon, which could have important implications for cancer therapy.

• The conclusions in our original Lancet paper need to be revised in the light of our further investigations.

• The observations are unchanged, but the underlying cause must be re-interpreted.

• Further studies of tumours using low b-values to measure perfusion are strongly recommended.