Intravoxel incoherent motion diffusion-weighted imaging in the liver: comparison of mono-, bi- and tri-exponential modelling at 3.0-T

Abstract

Purpose

To determine whether a mono-, bi- or tri-exponential model best fits the intravoxel incoherent motion (IVIM) diffusion-weighted imaging (DWI) signal of normal livers.

Materials and methods

The pilot and validation studies were conducted in 38 and 36 patients with normal livers, respectively. The DWI sequence was performed using single-shot echoplanar imaging with 11 (pilot study) and 16 (validation study) b values. In each study, data from all patients were used to model the IVIM signal of normal liver.

Diffusion coefficients (D_i ± standard deviations) and their fractions (f_i ± standard deviations) were determined from each model. The models were compared using the extra sum-of-squares test and information criteria.

Results

The tri-exponential model provided a better fit than both the bi- and mono-exponential models. The tri-exponential IVIM model determined three diffusion compartments: a slow (D₁ = 1.35 ± 0.03 × 10^-3 mm²/s; f₁ = 72.7 ± 0.9 %), a fast (D₂ = 26.50 ± 2.49 × 10^-3 mm²/s; f₂ = 13.7 ± 0.6 %) and a very fast (D₃ = 404.00 ± 43.7 × 10^-3 mm²/s; f₃ = 13.5 ± 0.8 %) diffusion compartment [results from the validation study]. The very fast compartment contributed to the IVIM signal only for b values ≤15 s/mm²

Conclusion

The tri-exponential model provided the best fit for IVIM signal decay in the liver over the 0-800 s/mm² range. In IVIM analysis of normal liver, a third very fast (pseudo)diffusion component might be relevant.

Key Points

• For normal liver, tri-exponential IVIM model might be superior to bi-exponential

• A very fast compartment (D = 404.00 ± 43.7 × 10 ^-3 mm² /s; f = 13.5 ± 0.8 %) is determined from the tri-exponential model

• The compartment contributes to the IVIM signal only for b ≤ 15 s/mm ²

Appendix 1

¹H-MR Spectroscopy

¹H-MRS was performed in all patients. Semiautomated optimization of gradient shimming followed by manual adjustment of central frequency was performed, and water line widths of less than 25Hz were obtained. Water suppression was not performed for any of the sequences.

Single-voxel MR spectroscopic data were acquired using seven breath-hold point-resolved spatially localized spectroscopic (PRESS) pulse sequences (repetition time, 5000 msec; three acquisitions; 2048 data points over 1250 Hz spectral width; and acquisition time, 15 sec) with echo times of 30, 40, 50, 60, 80, 100, and 135 msec to measure the T2 relaxation times of water and methylene (CH₂). A long TR was used to minimize T1 effects. Each breath-hold lasted 15 sec.

In all patients, the same 30*30*30-mm (27 ml) voxel was used for these seven sequences and was positioned obliquely on segment VII on the transverse low-T1-weighted section (thus avoiding extra-hepatic fat, large hepatic vessels, and organs adjacent to the liver).

¹H-MR spectroscopic data analysis

The Java-based MR user interface spectroscopic analysis package (jMRUI; A van den Boogaart, Catholic University, Leuven, Belgium) was used for the time-domain analysis. Metabolite signals were analyzed using the Advanced Magnetic Resonance (AMARES) fitting algorithm within jMRUI. We measured the water peak at 4.76 ppm and the methylene peak at 1.33 ppm. Spectra were used only if homogeneity after shimming, measured as the full width at 50 % peak height, was better than 25 Hz. Peak integrals were quantified by fitting to a Gaussian line shape.

T2 relaxation times of both metabolites were determined from their peak amplitudes at each echo time using an exponential least-squares fitting algorithm. The peak areas of the methylene and water signals were then corrected for T2 effects (i.e., theoretical peak areas with 0 echo time), using the individually calculated T2 relaxation times.

LFC was calculated as follows:

$$ {\mathrm{LFC}}_{\mathrm{MRS}} = 100*{\mathrm{A}}_{0\mathrm{msec}\_\mathrm{C}\mathrm{H}2}/\ \left({\mathrm{A}}_{0\mathrm{msec}\_\mathrm{C}\mathrm{H}2} + {\mathrm{A}}_{0\mathrm{msec}\_\mathrm{WATER}}\right) $$

where A_{0msec_CH2} and A_{0msec_WATER} were the areas of the methylene and water peaks, respectively, corrected for both T1 and T2 effects.

Triple-echo acquisition

A transverse breath-hold T1-weighted two-dimensional triple-echo spoiled gradient-echo sequence was performed through the liver with the following parameters: repetition time (msec)/echo time (msec) of 192/2.46 (in-phase [IP₁]), 3.69 (opposed-phase [OP]), and 4.92 (in-phase [IP₂]); flip angle, 60°; section thickness, 6 mm; intersection gap, 1.2 mm; matrix, 256*192; number of sections, 25; and acquisition time, 34 seconds. Parallel imaging (with an acceleration factor of 2) was performed using generalized autocalibrating partially parallel acquisition (GRAPPA; Siemens Medical Solutions Erlangen, Germany). Two separate breath-holds (each lasting 17 seconds) were needed to cover the entire liver volume.

Measurement of T2*

An ROI of 1–2 cm in diameter was drawn in the liver (avoiding large vessels, bile ducts or extra-hepatic areas) in five slices. The signal intensity (SI) in these five ROIs was averaged and recorded for IP1 (TE, 2.46 msec) and IP2 (TE, 4.92 msec) T1-weighted MR images. We used the copy-and-paste function of the workstation (Leonardo; Siemens Medical Solutions) to draw exactly the same ROIs at the same locations on IP images.

As previously reported [37, 38], the IP1 and IP2 images were used to estimate T2* relaxation time.