ORIGINAL ARTICLE
Impact of Measurement Parameters on Apparent Diffusion Coefficient Quantification in Diffusion-Weighted-Magnetic Resonance Imaging Holger Schmidt, PhD,* Sergios Gatidis, MD,* Nina F. Schwenzer, MD,* and Petros Martirosian, PhD† Objective: The scope of this work was to systematically evaluate the reproducibility of diffusion-weighted imaging and the impact of b values used for apparent diffusion coefficient (ADC) calculation as well as the echo time (TE) on the resulting ADC in phantom studies. We attempted to find a minimum upper b value needed for reliable ADC measurements. In addition, we were able to investigate these impacts not only for different diffusivities but also for different T2 relaxation times. The influence of different b values on ADC calculations for different organs was also assessed in a volunteer study. Materials and Methods: Diffusion-weighted imaging of a phantom consisting of 16 compartments with combinations of 4 different diffusivities and 4 different T2 relaxation times was conducted 5 times using 11 b values (0–1000 s/mm2) and 5 different TEs. Apparent diffusion coefficient was calculated from the 16 compartment regions of interest using 42 different combinations of b values. Reproducibility of ADC was assessed from the coefficient of variation of the 5 measurements. The ADC stability was determined from a voxel-based coefficient of variation (CVsta) and the signal-to-noise ratio (SNR) to find the minimum upper b values for a reliable ADC quantification. The influence of TE on ADC quantification was assessed for 9 different b value combinations. The influence of 9 different b value combinations on ADC was evaluated by a region of interest analysis of 7 organs in 12 volunteers. Results: The found coefficient of variation was between 10.2% and 1.4%, decreasing with increasing upper b value and increasing diffusivities. Accordingly, CVsta and SNR showed the same trend. Using an upper b value of 600 s/mm2 gives already reliable ADC results showing a maximum CVsta of 7.5%, whereas an upper b value of 1000 s/mm2 revealed a maximum CVsta of 5.5%. Values of ADC reduced with increasing upper b value in phantom as well as in human data. Apparent diffusion coefficient also reduced with increasing TE and tended to increase for increasing T2 relaxation times and increasing diffusion restriction. Conclusions: Apparent diffusion coefficient can be measured with high reproducibility but strongly depends on b values used and TE, which should be kept constant in each examination protocol. Whereas upper b values as low as 400 s/mm2 can be used for examinations of tissues with low diffusivities, very high b values (>1000 s/mm2) are needed to reach an optimal SNR for high diffusive tissues. An upper b value of 600 s/mm2 is a good compromise regarding ADC stability, SNR, and measurement time for all tissue types. Key Words: diffusion-weighted imaging, magnetic resonance imaging, quantification (Invest Radiol 2015;50: 46–56)
Received for publication June 12, 2014; and accepted for publication, after revision, August 6, 2014. From the *Department of Radiology, Diagnostic and Interventional Radiology, and †Department of Radiology, Diagnostic and Interventional Radiology, Section on Experimental Radiology, Eberhard Karls University, Tübingen, Germany. Conflicts of interest and sources of funding: none declared. Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.investigativeradiology.com). Reprints: Holger Schmidt, PhD, Department of Radiology, Diagnostic and Interventional Radiology, Eberhard Karls University, Hoppe-Seyler-Str. 3, 72076 Tübingen, Germany. E-mail: [email protected]. Copyright © 2014 by Lippincott Williams & Wilkins ISSN: 0020-9996/15/5001–0046
46
www.investigativeradiology.com
W
ith modern magnetic resonance imaging (MRI), whole-body diffusion-weighted imaging (DWI) is a valuable tool in tissue and lesion characterization. Diffusion-weighted imaging highlights lesions with high cellularity, a histological characteristic that is found in tumors with tightly packed cells such as neuroendocrine tumors, several sarcomas, and small cell cancers but also in breast cancer, lymphoma, or myeloma.1–3 Thus, the addition of DWI to the examination protocol has been shown to improve rates of lesion detection and diagnostic accuracy.4 However, signal intensities in diffusion-weighted images depend not only on diffusion restriction but also on tissue T2 relaxation times (eg, T2 shine-through or dark-through effects).5 However, quantitative analysis reflecting the diffusivity of the imaged tissue is possible using apparent diffusion coefficient (ADC) maps. Treatment-related changes in ADC values can be measured after successful chemotherapy and radiation therapy but also after radiofrequency ablation therapy.6,7 In liver metastases, the change in ADC has been shown to precede changes in lesions size.8 The ADC value could also play a role as a prognostic parameter in therapy monitoring of lung cancer.9 In recent years, substantial progress has been made in sequence design and clinical application of DWI. However, there are still challenges concerning quantification and reproducibility4 because ADCs can vary significantly between different measurements and scanners and ADC values of normal and pathologically altered tissues often overlap.4,10 In addition, it was shown that the assumption of monoexponential behavior of the DWI signal used for ADC calculation holds not true for some organs.11,12 Recently, a phantom allowing an independent adjustment of ADC values and T2 relaxation times based on aqueous solutions containing polyethylene-glycol (PEG) and gadobutrol was proposed.13 The goal of this work was to systematically investigate the influence of different diffusion weightings (b values) used for ADC calculation on the resulting ADC for different T2 relaxation times and diffusivities. In addition, the dependence of the echo time (TE) on the ADC was investigated. The non–monoexponential behavior of the DWI signal found in the phantom data was also identified in DWI data of 12 volunteers.
MATERIALS AND METHODS Phantom Studies Phantom measurements were performed on a 3-T clinical positron emission tomography/magnetic resonance scanner (Biograph mMR; Siemens Healthcare, Erlangen, Germany) using an MRI phantom for DWI recently described in the study of Gatidis et al.13 The phantom consists of 16 compartments arranged in 4 rows and 4 columns with independently adjusted ADC values (diffusivities D, approximately 2.2, 1.2, 0.8, 0.5 10−3 mm2/s; determined from ADC calculation using 11 b values listed below) and T2 relaxation times (T2, approximately 400, 200, 100, 50 milliseconds) on the basis of aqueous solutions containing PEG and gadobutrol (Fig. 1). Each DWI protocol was performed 5 times with new shimming procedure before each measurement. A single-shot spin-echo echo planar imaging sequence in 3-scan–trace mode with monopolar diffusion gradients, 11 b values (0, 100, 200, 300, 400, 500, 600, 700, Investigative Radiology • Volume 50, Number 1, January 2015
Copyright © 2014 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Investigative Radiology • Volume 50, Number 1, January 2015
Evaluation of ADC Quantification
measurements was obtained from the relative standard deviation (SD) of the mean ADC values from the 5 measurements (coefficient of variation): CVrep ¼ SD=ADCmean To evaluate the stability of ADC calculations using different b value combinations, a coefficient of variation (CVsta) was determined using the SD of the ADC from all voxels in each ROI and all 5 measurements. In addition, an SNR was estimated by dividing ADCmean of an ROI by the SD of the ADC within this ROI: SNR ¼ ADCmean =SD For the evaluation of the influence of ADC on different TEs, relative differences ΔADCTE were compared to mean ADC from measurements with TE of 52 milliseconds: ΔADCTE ¼ ðADCmean − ADCmean
TE¼52ms Þ=ADCmean TE ¼ 52 ms
Volunteer Study
FIGURE 1. Scheme of the diffusion phantom used for the phantom study. The phantom consists of 16 compartments with 4 different diffusivities and 4 different T2 relaxation times.
800, 900, and 1000 s/mm2), and the following parameters was used: TE, 52 milliseconds; repetition time, 5500 milliseconds; field of view, 450 192 mm; matrix size, 182 78; slice thickness, 2.5 mm; number of slices, 25; bandwidth, 2747 hertz per pixel; parallel imaging factor, 2; number of averages, 3; and spectral attenuated inversion recovery fat suppression. For the assessment of the influence of TE on the ADC calculation, DWI with same parameters but 5 different TEs (52, 62, 72, 82, and 92 milliseconds) was acquired 5 times each.
Volunteer Study For the volunteer study, the same scanner and sequence were used under free breathing using the following parameters: 11 b values (0, 100, 200, 300, 400, 500, 600, 700, 800, 900, and 1000 s/mm2); TE, 62 milliseconds; repetition time, 4300 milliseconds; field of view, 420 341 mm; matrix size, 192 156; slice thickness, 5 mm; number of slices, 20; bandwidth, 1860 hertz per pixel; parallel imaging factor, 2; number of averages, 5; and spectral attenuated inversion recovery fat suppression. A brain and abdominal examination was carried out for the 12 volunteers (8 males; age, 31 ± 8 years). The study was approved by the local ethics committee.
For the volunteer study, in addition to b0 values of 0 and 100 s/mm2, b0 = 200 and 300 s/mm2 were also used for the ADC calculation. A lower b value of 300 s/mm2 can be considered to be perfusionfree.14 Reference ADCs (ADCref) were calculated using all b values starting from the respective b0 (eg, b = 300, 400, 500, 600, 700, 800, 900, and 1000 s/mm2 for the b0 = 300 s/mm2 data).
TABLE 1. Overview of the b Value Combinations Used for ADC Calculation 2 b Values 2
b0, s/mm 0 0 0 0 0 0 0 0 0 0
3 b Values 2
b1, s/mm
b0, s/mm
b1, s/mm2
b2, s/mm2
100 200 300 400 500 600 700 800 900 1000
0 0 0 0 0 0 0 0 0 0 0 0
100 200 300 200 300 400 300 400 500 400 500 600
400 400 400 600 600 600 800 800 800 1000 1000 1000
100 100 100 100 100 100 100 100 100 100 100
200 300 200 300 400 300 400 500 400 500 600
400 400 600 600 600 800 800 800 1000 1000 1000
Data Evaluation The calculation of ADC values was carried out assuming a monoexponential model using MATLAB (The MathWorks Inc, Natick, MA). The b values used for the ADC computation are listed in Table 1. As reference value, ADCref calculated from all 11 b values (monoexponential model, b = 01000 s/mm2) was used. A Rician noise model was not applied because signal-to-noise ratio (SNR) was high enough for all data acquired (SNR, > > 2). The difference in ADC from each calculation with different b value combinations was calculated as follows: ΔADC ¼ ðADCmean − ADCref Þ=ADCref
Phantom Studies Mean ADC values (ADCmean) were derived from regions of interest (ROIs) of the 16 compartments. Reproducibility of the ADC © 2014 Lippincott Williams & Wilkins
100 200 100 300 100 400 100 500 100 600 100 700 100 800 100 900 100 1000 Combinations for b0 = 200 and 300 s/mm2 accordingly.
2
ADC indicates apparent diffusion coefficient.
www.investigativeradiology.com
Copyright © 2014 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
47
Investigative Radiology • Volume 50, Number 1, January 2015
Schmidt et al
Mean ADC values were derived from ROIs of kidney cortex, bone marrow (lumbar spine), muscle (autochthonous back musculature), right liver lobe, spleen, as well as gray matter and white matter. Care was taken to spare larger vessels and outer organ borders. The ROIs contained at least 50 image voxels; liver ROIs, more than 500 voxels. The ADC values were averaged over the 12 volunteers. In addition, intersubject variability was determined by calculating a coefficient of variation (CVpat) of perfusion-free ADCs (b0 = 300 s/mm2 and b1 = 400–1000 s/mm2). By calculating a 95% confidence interval (CIpat) (CIpat = 1.96 CVpat) and comparing the mean difference between ADCs calculated with an upper b value of b1 = 400 to 900 s/mm2 and ADC calculated with b1 = 1000 s/mm2, we could estimate which difference in ADC is significantly detectable within the intersubject variability with a confidence of 95%.
of ADC dependence on different b value combinations in the phantom and volunteer studies. For the examination of ADC dependence on TE, a 2-way ANOVA was used to account for both variations in TE and b value combinations. In addition, we performed a Bonferroni post hoc test to assess the different parameters. P values of less than 0.05 were considered as statistically significant. Statistical calculations were performed using Statistical Package for the Social Sciences (SPSS) (SPSS version 20; SPSS, Chicago, IL). The effect size was calculated manually using the formula η2 = sum of squareseffect/sum of squarestotal.
Statistics
Reproducibility of ADC for the 5 repeated measurements increased with increasing the upper b value used for the ADC calculation (CVrep[max] = 10.2% using b0 = 0 s/mm2 and b1 = 100 s/mm2,
To test for statistically significant differences between ADCs, a 1-way analysis of variance (ANOVA) was used for the investigation
RESULTS Phantom Studies Reproducibility
FIGURE 2. Coefficient of variation of the ADC from all ROIs of 5 measurements for all 16 compartments. Compartments with T2 = 400 milliseconds are shown in A; T2 = 200 milliseconds, in B; T2 = 100 milliseconds, in C; T2 = 50 milliseconds, in D. Cross: D = 0.5 10−3 mm2/s, circle: D = 0.8 10−3 mm2/s, triangle left: D = 1.2 10−3 mm2/s, triangle right: D = 2.2 10−3 mm2/s. First 10 data points: 2 b values, b0 = 0 s/mm2, b1 = 100 to 1000 s/mm2 (Table 1, upper left); second 9 data points: 2 b values, b0 = 100 s/mm2, b1 = 200 to 1000 s/mm2 (Table 1, lower left); third 12 data points: 3 b values, b0 = 0 s/mm2, b2 = 400, 600, 800, 1000 s/mm2 (Table 1, upper right); fourth 11 data points: 3 b values, b0 = 100 s/mm2, b2 = 400, 600, 800, 1000 s/mm2 (Table 1, lower right).
48
www.investigativeradiology.com
© 2014 Lippincott Williams & Wilkins
Copyright © 2014 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Investigative Radiology • Volume 50, Number 1, January 2015
CVrep[min] = 1.4% using b0 = 100 s/mm2 and b1 = 1000 s/mm2; see Supplemental Material 1, Supplemental Digital Content 1, http://links.lww.com/RLI/A173). Reproducibility tended to be higher for lower diffusion restrictions and showed no correlation with T2 relaxation times.
Stability Accordingly, the stability of ADC measurements also increased with increasing the upper b value used for ADC calculation (Fig. 2). For the high diffusion restrictions (D = 0.5 10−3 mm2/s, T2 = 400 milliseconds) calculated with 2 b values and b0 = 0 s/mm2, CVsta decreased from approximately 25% to 5% using an upper b value of b1 = 100 and b1 = 1000 s/mm2, respectively (Fig. 2A, crosses in the first data block). For the lower diffusion restrictions, the reduction was less prominent (CVsta reduction from 6% to 2.5% for the same b value combination and T2 but D = 2.2 10−3 mm2/s; Fig. 2A, triangles right in the first data block). The CVsta decreased with decreasing diffusion restriction (compare course of crosses, circles, and triangles in Fig. 2), whereas no correlation with T2 relaxation times was found. However, when using very small upper b values (eg, b1 = 100 or 200 s/mm2), compartments with higher T2 relaxation times tended to show larger CVsta (compare, eg, the first data points in the first data block in Fig. 2A with respective data points in Fig. 2, B–D).
Signal-to-Noise Ratio Correspondingly, SNR of ADC values increased with increasing the upper b value used for ADC calculation (supplemental material 2, Supplemental Digital Content 2, http://links.lww.com/RLI/A174). Also, SNR tended to be higher in compartments with low diffusion restriction, whereas no correlation with T2 relaxation rates could be found. The SNR of ADC maps using a larger lower b value (b0 = 100 s/mm2)
Evaluation of ADC Quantification
was approximately 1.5% lower than that using b0 = 0 s/mm2. A third middle b value had only minor effects on SD and SNR.
ADC Dependence on b Value Combinations Dependencies of ADC on different b value combinations used for the calculation are depicted in Figure 3. In general, ADC decreased with increasing upper b value. The ADC values were less dependent on the lower b value (except for the small upper b values, where ADC fitting was unstable). Also, the middle b values for the 3 b value calculations showed only minor effect on the resulting ADC. Small offsets in ADC values were visible between compartments of different T2 relaxation times but same diffusion restrictions for D = 1.2, 0.8, and 0.5 10−3 mm2/s. These were caused by small imperfections in diffusivity adjustments with PEG. A difference plot of the calculated ADCs to ADCref (ΔADC) is shown for a 2 b value case (b0 = 100 mm/s2) in Figure 4; the difference plots for the remaining b value combinations can be assessed in supplemental material 3, Supplemental Digital Content 3, http://links.lww.com/RLI/A175. In concordance with the stability results, highest ΔADCs were found for b0 = 0 and b1 = 100 s/mm2 in compartments with high diffusion restriction (up to −19%, see supplemental material 3A, Supplemental Digital Content 3, http://links.lww.com/RLI/A175). For different T2 relaxation rates, no dependencies on ADC differences could be found. The ANOVA analysis revealed that differences in the ADCs from different b value combinations within compartments are nearly all significant (Table 2), except for compartments with low diffusion restriction (D = 2.2 10−3 mm2/s) where differences remained very constant. For the remaining 3 diffusivities, differences increased with decreasing diffusion restriction. Effect sizes of the ADC dependencies on b value combinations show the same behavior and accounted mostly for more than 30% of the total ADC variation (Table 2). Higher
FIGURE 3. Mean ADC values and standard deviations from ROI analysis of the 16 compartments for 5 repeated measurements. Blue: T2 = 400 milliseconds, red: T2 = 200 milliseconds, green: T2 = 100 milliseconds, black: T2 = 50 milliseconds. Cross: D = 0.5 10−3 mm2/s, circle: D = 0.8 10−3 mm2/s, triangle left: D = 1.2 10−3 mm2/s, triangle right: D = 2.2 10−3 mm2/s. First 10 data points: 2 b values, b0 = 0 s/mm2, b1 = 100 to 1000 s/mm2 (Table 1, upper left); second 9 data points: 2 b values, b0 = 100 s/mm2, b1 = 200 to 1000 s/mm2 (Table 1, lower left); third 12 data points: 3 b values, b0 = 0 s/mm2, b2 = 400, 600, 800, 1000 s/mm2 (Table 1, upper right); fourth 11 data points: 3 b values, b0 = 100 s/mm2, b2 = 400, 600, 800, 1000 s/mm2 (Table 1, lower right).
© 2014 Lippincott Williams & Wilkins
www.investigativeradiology.com
Copyright © 2014 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
49
Investigative Radiology • Volume 50, Number 1, January 2015
Schmidt et al
FIGURE 4. Relative differences of ADCs to the reference ADC for calculations using 2 b values with b0 = 100 s/mm2. Blue: T2 = 400 milliseconds, red: T2 = 200 milliseconds, green: T2 = 100 milliseconds, black: T2 = 50 milliseconds. First block of each color corresponds to compartments with D = 0.5 10−3 mm2/s; second block, D = 0.8 10−3 mm2/s; third block, D = 1.2 10−3 mm2/s; and fourth block, D = 2.2 10−3 mm2/s. Each block consists of 9 data points corresponding to ADC calculated with b1 = 200 to 1000 s/mm2 (Table 1, lower left).
dependencies of ADC on b value combinations (higher significances and effect sizes) were found using a lower b value of b0 = 100 s/mm2 and using 2 instead of 3 b values for the ADC calculation. Scatter plots including a linear fit of all measurements for all compartments and
different diffusivities can be found in supplemental material 4, Supplemental Digital Content 4, http://links.lww.com/RLI/A176; detailed results from the ANOVA analysis and the post hoc test, in supplemental material 5, Supplemental Digital Content 5, http://links.lww.com/RLI/A177.
TABLE 2. P Values and Effect Sizes (η2) From the 1-Way ANOVA of the ADC Differences Within the 16 Phantom Compartments Using Different b Value Combinations for ADC Calculation 2 b Values b0 = 0 s/mm2
Compartments −3
D, 10 0.5 0.8 1.2 2.2 0.5 0.8 1.2 2.2 0.5 0.8 1.2 2.2 0.5 0.8 1.2 2.2
2
mm /s T2, milliseconds 400 400 400 400 200 200 200 200 100 100 100 100 50 50 50 50
2 b Values b0 = 100 s/mm2
3 b Values b0 = 0 s/mm2
3 b Values b0 = 100 s/mm2
P
η
P
η
P
η
P
η2
0.0049 0.0003
Impact of Measurement Parameters on Apparent Diffusion Coefficient Quantification in Diffusion-Weighted-Magnetic Resonance Imaging Holger Schmidt, PhD,* Sergios Gatidis, MD,* Nina F. Schwenzer, MD,* and Petros Martirosian, PhD† Objective: The scope of this work was to systematically evaluate the reproducibility of diffusion-weighted imaging and the impact of b values used for apparent diffusion coefficient (ADC) calculation as well as the echo time (TE) on the resulting ADC in phantom studies. We attempted to find a minimum upper b value needed for reliable ADC measurements. In addition, we were able to investigate these impacts not only for different diffusivities but also for different T2 relaxation times. The influence of different b values on ADC calculations for different organs was also assessed in a volunteer study. Materials and Methods: Diffusion-weighted imaging of a phantom consisting of 16 compartments with combinations of 4 different diffusivities and 4 different T2 relaxation times was conducted 5 times using 11 b values (0–1000 s/mm2) and 5 different TEs. Apparent diffusion coefficient was calculated from the 16 compartment regions of interest using 42 different combinations of b values. Reproducibility of ADC was assessed from the coefficient of variation of the 5 measurements. The ADC stability was determined from a voxel-based coefficient of variation (CVsta) and the signal-to-noise ratio (SNR) to find the minimum upper b values for a reliable ADC quantification. The influence of TE on ADC quantification was assessed for 9 different b value combinations. The influence of 9 different b value combinations on ADC was evaluated by a region of interest analysis of 7 organs in 12 volunteers. Results: The found coefficient of variation was between 10.2% and 1.4%, decreasing with increasing upper b value and increasing diffusivities. Accordingly, CVsta and SNR showed the same trend. Using an upper b value of 600 s/mm2 gives already reliable ADC results showing a maximum CVsta of 7.5%, whereas an upper b value of 1000 s/mm2 revealed a maximum CVsta of 5.5%. Values of ADC reduced with increasing upper b value in phantom as well as in human data. Apparent diffusion coefficient also reduced with increasing TE and tended to increase for increasing T2 relaxation times and increasing diffusion restriction. Conclusions: Apparent diffusion coefficient can be measured with high reproducibility but strongly depends on b values used and TE, which should be kept constant in each examination protocol. Whereas upper b values as low as 400 s/mm2 can be used for examinations of tissues with low diffusivities, very high b values (>1000 s/mm2) are needed to reach an optimal SNR for high diffusive tissues. An upper b value of 600 s/mm2 is a good compromise regarding ADC stability, SNR, and measurement time for all tissue types. Key Words: diffusion-weighted imaging, magnetic resonance imaging, quantification (Invest Radiol 2015;50: 46–56)
Received for publication June 12, 2014; and accepted for publication, after revision, August 6, 2014. From the *Department of Radiology, Diagnostic and Interventional Radiology, and †Department of Radiology, Diagnostic and Interventional Radiology, Section on Experimental Radiology, Eberhard Karls University, Tübingen, Germany. Conflicts of interest and sources of funding: none declared. Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.investigativeradiology.com). Reprints: Holger Schmidt, PhD, Department of Radiology, Diagnostic and Interventional Radiology, Eberhard Karls University, Hoppe-Seyler-Str. 3, 72076 Tübingen, Germany. E-mail: [email protected]. Copyright © 2014 by Lippincott Williams & Wilkins ISSN: 0020-9996/15/5001–0046
46
www.investigativeradiology.com
W
ith modern magnetic resonance imaging (MRI), whole-body diffusion-weighted imaging (DWI) is a valuable tool in tissue and lesion characterization. Diffusion-weighted imaging highlights lesions with high cellularity, a histological characteristic that is found in tumors with tightly packed cells such as neuroendocrine tumors, several sarcomas, and small cell cancers but also in breast cancer, lymphoma, or myeloma.1–3 Thus, the addition of DWI to the examination protocol has been shown to improve rates of lesion detection and diagnostic accuracy.4 However, signal intensities in diffusion-weighted images depend not only on diffusion restriction but also on tissue T2 relaxation times (eg, T2 shine-through or dark-through effects).5 However, quantitative analysis reflecting the diffusivity of the imaged tissue is possible using apparent diffusion coefficient (ADC) maps. Treatment-related changes in ADC values can be measured after successful chemotherapy and radiation therapy but also after radiofrequency ablation therapy.6,7 In liver metastases, the change in ADC has been shown to precede changes in lesions size.8 The ADC value could also play a role as a prognostic parameter in therapy monitoring of lung cancer.9 In recent years, substantial progress has been made in sequence design and clinical application of DWI. However, there are still challenges concerning quantification and reproducibility4 because ADCs can vary significantly between different measurements and scanners and ADC values of normal and pathologically altered tissues often overlap.4,10 In addition, it was shown that the assumption of monoexponential behavior of the DWI signal used for ADC calculation holds not true for some organs.11,12 Recently, a phantom allowing an independent adjustment of ADC values and T2 relaxation times based on aqueous solutions containing polyethylene-glycol (PEG) and gadobutrol was proposed.13 The goal of this work was to systematically investigate the influence of different diffusion weightings (b values) used for ADC calculation on the resulting ADC for different T2 relaxation times and diffusivities. In addition, the dependence of the echo time (TE) on the ADC was investigated. The non–monoexponential behavior of the DWI signal found in the phantom data was also identified in DWI data of 12 volunteers.
MATERIALS AND METHODS Phantom Studies Phantom measurements were performed on a 3-T clinical positron emission tomography/magnetic resonance scanner (Biograph mMR; Siemens Healthcare, Erlangen, Germany) using an MRI phantom for DWI recently described in the study of Gatidis et al.13 The phantom consists of 16 compartments arranged in 4 rows and 4 columns with independently adjusted ADC values (diffusivities D, approximately 2.2, 1.2, 0.8, 0.5 10−3 mm2/s; determined from ADC calculation using 11 b values listed below) and T2 relaxation times (T2, approximately 400, 200, 100, 50 milliseconds) on the basis of aqueous solutions containing PEG and gadobutrol (Fig. 1). Each DWI protocol was performed 5 times with new shimming procedure before each measurement. A single-shot spin-echo echo planar imaging sequence in 3-scan–trace mode with monopolar diffusion gradients, 11 b values (0, 100, 200, 300, 400, 500, 600, 700, Investigative Radiology • Volume 50, Number 1, January 2015
Copyright © 2014 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Investigative Radiology • Volume 50, Number 1, January 2015
Evaluation of ADC Quantification
measurements was obtained from the relative standard deviation (SD) of the mean ADC values from the 5 measurements (coefficient of variation): CVrep ¼ SD=ADCmean To evaluate the stability of ADC calculations using different b value combinations, a coefficient of variation (CVsta) was determined using the SD of the ADC from all voxels in each ROI and all 5 measurements. In addition, an SNR was estimated by dividing ADCmean of an ROI by the SD of the ADC within this ROI: SNR ¼ ADCmean =SD For the evaluation of the influence of ADC on different TEs, relative differences ΔADCTE were compared to mean ADC from measurements with TE of 52 milliseconds: ΔADCTE ¼ ðADCmean − ADCmean
TE¼52ms Þ=ADCmean TE ¼ 52 ms
Volunteer Study
FIGURE 1. Scheme of the diffusion phantom used for the phantom study. The phantom consists of 16 compartments with 4 different diffusivities and 4 different T2 relaxation times.
800, 900, and 1000 s/mm2), and the following parameters was used: TE, 52 milliseconds; repetition time, 5500 milliseconds; field of view, 450 192 mm; matrix size, 182 78; slice thickness, 2.5 mm; number of slices, 25; bandwidth, 2747 hertz per pixel; parallel imaging factor, 2; number of averages, 3; and spectral attenuated inversion recovery fat suppression. For the assessment of the influence of TE on the ADC calculation, DWI with same parameters but 5 different TEs (52, 62, 72, 82, and 92 milliseconds) was acquired 5 times each.
Volunteer Study For the volunteer study, the same scanner and sequence were used under free breathing using the following parameters: 11 b values (0, 100, 200, 300, 400, 500, 600, 700, 800, 900, and 1000 s/mm2); TE, 62 milliseconds; repetition time, 4300 milliseconds; field of view, 420 341 mm; matrix size, 192 156; slice thickness, 5 mm; number of slices, 20; bandwidth, 1860 hertz per pixel; parallel imaging factor, 2; number of averages, 5; and spectral attenuated inversion recovery fat suppression. A brain and abdominal examination was carried out for the 12 volunteers (8 males; age, 31 ± 8 years). The study was approved by the local ethics committee.
For the volunteer study, in addition to b0 values of 0 and 100 s/mm2, b0 = 200 and 300 s/mm2 were also used for the ADC calculation. A lower b value of 300 s/mm2 can be considered to be perfusionfree.14 Reference ADCs (ADCref) were calculated using all b values starting from the respective b0 (eg, b = 300, 400, 500, 600, 700, 800, 900, and 1000 s/mm2 for the b0 = 300 s/mm2 data).
TABLE 1. Overview of the b Value Combinations Used for ADC Calculation 2 b Values 2
b0, s/mm 0 0 0 0 0 0 0 0 0 0
3 b Values 2
b1, s/mm
b0, s/mm
b1, s/mm2
b2, s/mm2
100 200 300 400 500 600 700 800 900 1000
0 0 0 0 0 0 0 0 0 0 0 0
100 200 300 200 300 400 300 400 500 400 500 600
400 400 400 600 600 600 800 800 800 1000 1000 1000
100 100 100 100 100 100 100 100 100 100 100
200 300 200 300 400 300 400 500 400 500 600
400 400 600 600 600 800 800 800 1000 1000 1000
Data Evaluation The calculation of ADC values was carried out assuming a monoexponential model using MATLAB (The MathWorks Inc, Natick, MA). The b values used for the ADC computation are listed in Table 1. As reference value, ADCref calculated from all 11 b values (monoexponential model, b = 01000 s/mm2) was used. A Rician noise model was not applied because signal-to-noise ratio (SNR) was high enough for all data acquired (SNR, > > 2). The difference in ADC from each calculation with different b value combinations was calculated as follows: ΔADC ¼ ðADCmean − ADCref Þ=ADCref
Phantom Studies Mean ADC values (ADCmean) were derived from regions of interest (ROIs) of the 16 compartments. Reproducibility of the ADC © 2014 Lippincott Williams & Wilkins
100 200 100 300 100 400 100 500 100 600 100 700 100 800 100 900 100 1000 Combinations for b0 = 200 and 300 s/mm2 accordingly.
2
ADC indicates apparent diffusion coefficient.
www.investigativeradiology.com
Copyright © 2014 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
47
Investigative Radiology • Volume 50, Number 1, January 2015
Schmidt et al
Mean ADC values were derived from ROIs of kidney cortex, bone marrow (lumbar spine), muscle (autochthonous back musculature), right liver lobe, spleen, as well as gray matter and white matter. Care was taken to spare larger vessels and outer organ borders. The ROIs contained at least 50 image voxels; liver ROIs, more than 500 voxels. The ADC values were averaged over the 12 volunteers. In addition, intersubject variability was determined by calculating a coefficient of variation (CVpat) of perfusion-free ADCs (b0 = 300 s/mm2 and b1 = 400–1000 s/mm2). By calculating a 95% confidence interval (CIpat) (CIpat = 1.96 CVpat) and comparing the mean difference between ADCs calculated with an upper b value of b1 = 400 to 900 s/mm2 and ADC calculated with b1 = 1000 s/mm2, we could estimate which difference in ADC is significantly detectable within the intersubject variability with a confidence of 95%.
of ADC dependence on different b value combinations in the phantom and volunteer studies. For the examination of ADC dependence on TE, a 2-way ANOVA was used to account for both variations in TE and b value combinations. In addition, we performed a Bonferroni post hoc test to assess the different parameters. P values of less than 0.05 were considered as statistically significant. Statistical calculations were performed using Statistical Package for the Social Sciences (SPSS) (SPSS version 20; SPSS, Chicago, IL). The effect size was calculated manually using the formula η2 = sum of squareseffect/sum of squarestotal.
Statistics
Reproducibility of ADC for the 5 repeated measurements increased with increasing the upper b value used for the ADC calculation (CVrep[max] = 10.2% using b0 = 0 s/mm2 and b1 = 100 s/mm2,
To test for statistically significant differences between ADCs, a 1-way analysis of variance (ANOVA) was used for the investigation
RESULTS Phantom Studies Reproducibility
FIGURE 2. Coefficient of variation of the ADC from all ROIs of 5 measurements for all 16 compartments. Compartments with T2 = 400 milliseconds are shown in A; T2 = 200 milliseconds, in B; T2 = 100 milliseconds, in C; T2 = 50 milliseconds, in D. Cross: D = 0.5 10−3 mm2/s, circle: D = 0.8 10−3 mm2/s, triangle left: D = 1.2 10−3 mm2/s, triangle right: D = 2.2 10−3 mm2/s. First 10 data points: 2 b values, b0 = 0 s/mm2, b1 = 100 to 1000 s/mm2 (Table 1, upper left); second 9 data points: 2 b values, b0 = 100 s/mm2, b1 = 200 to 1000 s/mm2 (Table 1, lower left); third 12 data points: 3 b values, b0 = 0 s/mm2, b2 = 400, 600, 800, 1000 s/mm2 (Table 1, upper right); fourth 11 data points: 3 b values, b0 = 100 s/mm2, b2 = 400, 600, 800, 1000 s/mm2 (Table 1, lower right).
48
www.investigativeradiology.com
© 2014 Lippincott Williams & Wilkins
Copyright © 2014 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Investigative Radiology • Volume 50, Number 1, January 2015
CVrep[min] = 1.4% using b0 = 100 s/mm2 and b1 = 1000 s/mm2; see Supplemental Material 1, Supplemental Digital Content 1, http://links.lww.com/RLI/A173). Reproducibility tended to be higher for lower diffusion restrictions and showed no correlation with T2 relaxation times.
Stability Accordingly, the stability of ADC measurements also increased with increasing the upper b value used for ADC calculation (Fig. 2). For the high diffusion restrictions (D = 0.5 10−3 mm2/s, T2 = 400 milliseconds) calculated with 2 b values and b0 = 0 s/mm2, CVsta decreased from approximately 25% to 5% using an upper b value of b1 = 100 and b1 = 1000 s/mm2, respectively (Fig. 2A, crosses in the first data block). For the lower diffusion restrictions, the reduction was less prominent (CVsta reduction from 6% to 2.5% for the same b value combination and T2 but D = 2.2 10−3 mm2/s; Fig. 2A, triangles right in the first data block). The CVsta decreased with decreasing diffusion restriction (compare course of crosses, circles, and triangles in Fig. 2), whereas no correlation with T2 relaxation times was found. However, when using very small upper b values (eg, b1 = 100 or 200 s/mm2), compartments with higher T2 relaxation times tended to show larger CVsta (compare, eg, the first data points in the first data block in Fig. 2A with respective data points in Fig. 2, B–D).
Signal-to-Noise Ratio Correspondingly, SNR of ADC values increased with increasing the upper b value used for ADC calculation (supplemental material 2, Supplemental Digital Content 2, http://links.lww.com/RLI/A174). Also, SNR tended to be higher in compartments with low diffusion restriction, whereas no correlation with T2 relaxation rates could be found. The SNR of ADC maps using a larger lower b value (b0 = 100 s/mm2)
Evaluation of ADC Quantification
was approximately 1.5% lower than that using b0 = 0 s/mm2. A third middle b value had only minor effects on SD and SNR.
ADC Dependence on b Value Combinations Dependencies of ADC on different b value combinations used for the calculation are depicted in Figure 3. In general, ADC decreased with increasing upper b value. The ADC values were less dependent on the lower b value (except for the small upper b values, where ADC fitting was unstable). Also, the middle b values for the 3 b value calculations showed only minor effect on the resulting ADC. Small offsets in ADC values were visible between compartments of different T2 relaxation times but same diffusion restrictions for D = 1.2, 0.8, and 0.5 10−3 mm2/s. These were caused by small imperfections in diffusivity adjustments with PEG. A difference plot of the calculated ADCs to ADCref (ΔADC) is shown for a 2 b value case (b0 = 100 mm/s2) in Figure 4; the difference plots for the remaining b value combinations can be assessed in supplemental material 3, Supplemental Digital Content 3, http://links.lww.com/RLI/A175. In concordance with the stability results, highest ΔADCs were found for b0 = 0 and b1 = 100 s/mm2 in compartments with high diffusion restriction (up to −19%, see supplemental material 3A, Supplemental Digital Content 3, http://links.lww.com/RLI/A175). For different T2 relaxation rates, no dependencies on ADC differences could be found. The ANOVA analysis revealed that differences in the ADCs from different b value combinations within compartments are nearly all significant (Table 2), except for compartments with low diffusion restriction (D = 2.2 10−3 mm2/s) where differences remained very constant. For the remaining 3 diffusivities, differences increased with decreasing diffusion restriction. Effect sizes of the ADC dependencies on b value combinations show the same behavior and accounted mostly for more than 30% of the total ADC variation (Table 2). Higher
FIGURE 3. Mean ADC values and standard deviations from ROI analysis of the 16 compartments for 5 repeated measurements. Blue: T2 = 400 milliseconds, red: T2 = 200 milliseconds, green: T2 = 100 milliseconds, black: T2 = 50 milliseconds. Cross: D = 0.5 10−3 mm2/s, circle: D = 0.8 10−3 mm2/s, triangle left: D = 1.2 10−3 mm2/s, triangle right: D = 2.2 10−3 mm2/s. First 10 data points: 2 b values, b0 = 0 s/mm2, b1 = 100 to 1000 s/mm2 (Table 1, upper left); second 9 data points: 2 b values, b0 = 100 s/mm2, b1 = 200 to 1000 s/mm2 (Table 1, lower left); third 12 data points: 3 b values, b0 = 0 s/mm2, b2 = 400, 600, 800, 1000 s/mm2 (Table 1, upper right); fourth 11 data points: 3 b values, b0 = 100 s/mm2, b2 = 400, 600, 800, 1000 s/mm2 (Table 1, lower right).
© 2014 Lippincott Williams & Wilkins
www.investigativeradiology.com
Copyright © 2014 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
49
Investigative Radiology • Volume 50, Number 1, January 2015
Schmidt et al
FIGURE 4. Relative differences of ADCs to the reference ADC for calculations using 2 b values with b0 = 100 s/mm2. Blue: T2 = 400 milliseconds, red: T2 = 200 milliseconds, green: T2 = 100 milliseconds, black: T2 = 50 milliseconds. First block of each color corresponds to compartments with D = 0.5 10−3 mm2/s; second block, D = 0.8 10−3 mm2/s; third block, D = 1.2 10−3 mm2/s; and fourth block, D = 2.2 10−3 mm2/s. Each block consists of 9 data points corresponding to ADC calculated with b1 = 200 to 1000 s/mm2 (Table 1, lower left).
dependencies of ADC on b value combinations (higher significances and effect sizes) were found using a lower b value of b0 = 100 s/mm2 and using 2 instead of 3 b values for the ADC calculation. Scatter plots including a linear fit of all measurements for all compartments and
different diffusivities can be found in supplemental material 4, Supplemental Digital Content 4, http://links.lww.com/RLI/A176; detailed results from the ANOVA analysis and the post hoc test, in supplemental material 5, Supplemental Digital Content 5, http://links.lww.com/RLI/A177.
TABLE 2. P Values and Effect Sizes (η2) From the 1-Way ANOVA of the ADC Differences Within the 16 Phantom Compartments Using Different b Value Combinations for ADC Calculation 2 b Values b0 = 0 s/mm2
Compartments −3
D, 10 0.5 0.8 1.2 2.2 0.5 0.8 1.2 2.2 0.5 0.8 1.2 2.2 0.5 0.8 1.2 2.2
2
mm /s T2, milliseconds 400 400 400 400 200 200 200 200 100 100 100 100 50 50 50 50
2 b Values b0 = 100 s/mm2
3 b Values b0 = 0 s/mm2
3 b Values b0 = 100 s/mm2
P
η
P
η
P
η
P
η2
0.0049 0.0003
