•
Hypothesis & Rationale
Controversy exists in the specificity of dynamic contrast enhanced (DCE) MRI.
We have previously shown (Aref et al, Invest Radiol. 2002. 37:178-92) that the
spatial resolution greatly affects the values of contrast agent transfer rates
calculated from DCE MRI data using a two-compartment model. In DCE MRI
spatial resolution is sacrificed for temporal resolution. We and others (Su et al,
JMRI. 2003. 18:467-77) have hypothesized that this partial volume effect is the
reason for the controversy as different groups use vastly different spatial and
temporal resolutions. As a possible means of increasing temporal resolution
without sacrificing spatial resolution, many investigators (Kucharczyk et al, AJR.
1994. 163:671-9, Liang et al, IEEE Trans Med Imaging. 2003. 22:1026-30) have
tried using reduced encoding techniques. Below we analyze the limits of spatial
resolution and three reduced encoding techniques on quantitative
measurements of contrast agent transfer rates.
• We are testing the hypothesis that dynamic reference sets in reduced-encoding
techniques have spatial resolution limits for accurate quantitative tumor typing
based on volume normalized contrast agent transfer rates between tumor
plasma and extravascular extracellular space (EES), Kpt/VT, obtained from
dynamic contrast enhanced (DCE) MRI.

Plasma compartment equation
Materials & Methods
• Thirty-six 30-day old female
Sprague-Dawley rats were
injected with n-ethyl-n-nitrosourea.
Ten of these animals with
infiltrating ductal carcinomas were
analyzed in this study.
• Two-compartment model with Kp↔t
 PS (F >> PS)
Plasma compartment curve
Distribution
The two-compartment
model (above) and a Excretion
plot of the plasma
Injection
compartment (right).
Tumor compartment equation

Materials & Methods
Heart
Liver
Tumor Image
Ref.
Bladder

Results
FFT Keyhole Difference Images FFT TRIGR Difference Images
128 64 32 24 16 4 128 64 32 24 16 4
FFT RIGR Difference Images
• Keyhole shows “blurring” in the PE
direction, due to averaging.
• RIGR is more robust, but the
difference images have quality
losses with decreasing dynamic
data PE.
• TRIGR has qualitatively better
difference images with fewer
128 64 32 24 16 4 dynamic data PE, than Keyhole or
RIGR.

Results
PEDYN
Keyhole
RIGR
TRIGR
64
0.0016
0.0016
0.053
32
0.00
0.00
0.0055
24
0.00
0.00
0.053
16
0.00
0.00
0.00047
4 0.00
0.00
0.00047
Table 3: The p-value of the t-test adjusted step
down Bonferroni for Kp↔t/VT obtained from the
top five FFT “hot spot” locations in Keyhole,
RIGR, and TRIGR reconstructed with PEDYN =
128, 64, 32, 24, 16, and 4 compared to fully
reconstructed FFT (PE = 128) (n = 10).

Conclusions
• Keyhole: the top five Kpt/VT “hot spots” from fully reconstructed FFT and
Keyhole are statistically the same only at PEDYN = 128 and 64.
• RIGR: the top five Kpt/VT “hot spots” are statistically the same as FFT for all
PEDYN, as a result of very large standard deviations.
• TRIGR: the top five Kpt/VT “hot spots” from TRIGR and FFT are the same at
PEDYN = 128, 64, 32, and 24. In addition, using reduced dynamic data, PEDYN =
64 and 24, TRIGR localizes statistically similar “hot spots” as those obtained
from FFT maps.
• As PEDYN decreases the generalized-series techniques are better able to
accurately estimate image data and hence provide more accurate quantitative
dynamic contrast information than Keyhole.
• Although RIGR agrees with FFT at all PEDYN and appears statistically superior to
TRIGR, RIGR has unrealistic outlier Kpt/VT “hot spots” and large standard
deviations at lower PEDYN not seen with TRIGR. Furthermore, RIGR does not
localize statistically similar “hot spots” as FFT.
• Keyhole has the most limited dynamic data threshold and TRIGR more
accurately obtains clinical low-resolution dynamic data, PEDYN = 64, 32, and 24,
and more accurately localizes “hot spots”, PEDYN = 64 and 24, than both Keyhole
and RIGR.
• This implies that one can gain at least a fivefold (5.33) improvement in spatial
resolution without sacrificing the necessary temporal resolution.