Patients
This prospective study was conducted under the approval of the local institutional research ethics committee and informed consent was obtained from each patient. From February 2018 to August 2018, 41 patients who satisfied the inclusion criteria underwent customized MRI examination for this study. Inclusion criteria were that the patient must be diagnosed with colorectal liver metastasis and that the patient did not previously receive any local or systemic chemotherapy or radiotherapy. Eight patients were excluded for the following reasons: i) unsatisfactory tumor position, for example, the top-left liver lobe where the DW images usually present artifacts due to cardiac motion, (n = 1), ii) poor image quality that interfered with the delineation of tumors (n = 3), or iii) loss of follow-up (n = 4). Thus, 33 patients, including 19 males (a mean age of 56.7 years; range from 37 to 77 years) and 14 females (a mean age of 54.6 years; range from 39 to 67 years) were finally enrolled in this study.
The diagnosis of 20 patients was proved by biopsy or surgical resection. For the rest of patients (n = 13), the diagnosis was were established by typical MRI features of liver metastasis: 1) irregular or ill-defined borders with low T1 signal intensity (SI) and variable high T2 SI; 2) peripheral rim enhancement; 3) relatively hypo-enhancement on portal or delayed phase in comparison with liver parenchyma; 4) interval growth of at least 20% in the longest axial diameter on serial cross-sectional imaging.
All enrolled patients received a 2-week cycle of administration of CAPEOX and were treated with four to six cycles according to their treatment response and physical situation. CAPEOX was consisted of a 2-h intravenous infusion of oxaliplatin 130 mg/m2 on day 1 plus oral administration of capecitabine 1000 mg/m2 twice daily for 14 days. Baseline of MRI was performed on 1 day before start of the first cycle of chemotherapy. Posttreatment MRI was performed 4~6 days after completion of the first cycle.
MRI examination
All MRI examinations were conducted on a 3.0 T clinical whole-body MR imaging system (Ingenia, Philips Medical Systems, Eindhoven, The Netherlands) with a 32-channel phased-array coil. Patients were asked to fast for more than 5 h before scanning.
The scan protocol was consisted of the routine MRI sequences and a DWI sequence.
The routine MRI sequences included (a) an axial fat-suppressed T2-weighted turbo spin echo (TSE) sequence (TR/ TE 535/75 ms, slice thickness 7 mm, slice gap 1 mm, matrix size 232 × 199, FOV 350 mm × 392 mm), (b) coronal T2-weighted TSE sequence (TR/TE 1100/80 ms, slice thickness 6 mm, slice gap 1 mm, matrix size 292 × 253, FOV 350 mm × 346 mm), and (c) axial breath-hold dual-echo T1-weighted fast field-echo sequence (TR/TE1/TE2 106/1.15/2.3 ms, slice thickness 7 mm, slice gap 1 mm, matrix size 244 × 181, FOV 400 mm × 322 mm).
The DWI images were acquired using a spin-echo EPI sequence with gated-navigator respiratory motion compensation technique (the navigator gating window was set as 5 mm). Six b values (0, 200, 500, 1000, 1500, 2000 s/mm2 with 1, 1, 2, 2, 2, 4 signal averages, respectively) were applied in three orthogonal diffusion encoding directions [25]. The imaging parameters were: TR/TE 2000 /70 ms, number of slices 24, slice thickness 7 mm, slice gap 0 mm, matrix size 148 × 148, FOV 400 mm × 400 mm, NSA 2. The total scan time of the routine MRI sequences and DWI sequence was 24~27 min for different patients. Additionally, the patients who did not have pathologically confirmed hepatic metastases underwent a dynamic enhanced MR examination before the DWI scan for diagnosis.
The scan protocol was consisted of above routine MR sequences and a dynamic 3D breath-hold T1-weighted sequence with fat suppression. The dynamic enhanced MR imaging parameters were as follows: TR/TE1/TE2 3.6/1.32/2.3 ms, slice thickness 5 mm, slice gap 2.5 mm, matrix size 200 × 250, FOV 320 mm × 427 mm. The arterial, portal venous equilibrium and delayed phase were set as 20, 55, 90, and 180 s after contrast media injection, respectively.
Image analysis
All DW images obtained at pre-treatment and post-treatment (i.e. 4~6 days after completion of the first chemotherapy cycle) were independently reviewed by two radiologists with 15 and 7 years’ experience in reading MR images, respectively. They were all blinded to the final therapeutic response. For the pretreatment images, the largest tumor area slice of each patient was selected. On the slice, one or more regions of interest (ROIs) (if existing metastases) were created manually by each observer to cover as much of the solid part of tumors as possible and to avoid surrounding necrosis, hemorrhagic areas, vessels or bile ducts. As for the images after treatment, the slices with the same position of those before treatment were drawn in same way.
The ROIs were initially delineated on b0 image, then copied and pasted on corresponding DW images. The DW images were normalized by dividing them by the b0 images, and geometric mean of these images over different diffusion encoding directions was taken [25]. For each patient, the voxel-level DW data on ROIs were fitted by the non-mono-exponential diffusion models and the median values of calculated parameters over all ROIs were recorded as the measurements of the patient.
Above measurements were executed twice by observer I and were executed once by observer II regardless of pretreatment or posttreatment. The two executions for observer I was to evaluate intra-observer reproducibility. The assessment of inter-observer reproducibility was performed between the first measurement of observer I and the measurement of observer II. For each patient, the average of first measurements record by observer I and measurements recorded by observer II was calculated and taken as the final measurements of the patient.
The non-mono-exponential diffusion models are as follows:
-
1.
Kurtosis model [20]
$$ \frac{S(b)}{S_0}={e}^{\left(-{bADC}_k+\frac{1}{6}{b}^2{ADC}_k^2K\right)} $$
(1)
-
2.
Stretched exponential model [21]
$$ \frac{S(b)}{S_0}={e}^{-{(bDDC)}^{\alpha }} $$
(2)
-
3.
Statistical model [22]
$$ \frac{S(b)}{S_0}=\left(\frac{1+\phi \left(\frac{D_s}{\sigma \sqrt{2}}+\frac{b\sigma}{\sqrt{2}}\right)}{1+\phi \left(\frac{D_s}{\sigma \sqrt{2}}\right)}\right){e}^{\left(-{b D}_s+\frac{1}{2}{b}^2{\sigma}^2\right)} $$
(3)
where S(b) is the signal intensity for a givenvalue, S0 is S (b = 0 s/mm2), ADCk and K stands for respectively the apparent diffusion coefficient and the kurtosis coefficient of the kurtosis model, DDC and α is the distributed diffusion coefficient and the anomalous exponent term of the stretched exponential model, Ds andσ represents respectively the peak position and the width of diffusion coefficient distribution in the statistical model. The nonlinear regression is implemented using the Levenberg–Marquardt algorithm in Matlab (version R2014a, Mathworks, Natick, MA).
Percentage changes in the diffusion-related parameters (i.e. ADCk, K, DDC, α, Ds and σ) were calculated according to, where Pre-para and Post-para refer to the values of parameters before and after treatment, respectively.
Response evaluation
The response evaluation was performed on computed tomography (CT) between baseline and after 3 months from the start of the first cycle of chemotherapy. The overall response was determined according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 [26]. For patients with multiple metastases, the diameters of a maximum of two metastases in the liver were measured and summed to compare the total of those lesions’ diameters at baseline and after 3 months from the start of the first cycle of chemotherapy. Complete response (CR): no residual tumor was observed; partial response (PR): at least a 30% decrease was observed in the sum of the longest lesion diameters in comparison with the primary tumor size; progressive disease (PD): 20% or more increase in the sum of the longest diameters; and stable disease (SD): neither sufficient shrinkage to qualify for PR nor sufficient increase to qualify for PD. Patients who presented CR or PR were classified as responders, and patients who presented SD or PD were classified as non-responders.
Statistical analysis
All parameter values were expressed as means ± standard deviations. The difference of measured parameter values between intra- or inter-observers was assessed using the Mann-Whitney U test and the intraclass correlation coefficient (ICC) was calculated to evaluate intra- and inter-observer reproducibility. The repeatability of parameters was defined as excellent when ICC values were between 0.75~1, good when ICC values were between 0.60~0.74, fair when ICC values were between 0.40~0.59, and poor when ICC values were smaller than 0.4 [27]. The differences above diffusion-related parameters and their percentage changes before and after chemotherapy between response and non-response groups were assessed using independent-samples T-test (normality) and Mann–Whitney U-test (nonnormality). The diagnostic accuracy of indicators (Pre-K, ΔK, Post-σ, and Δσ) was evaluated in terms of sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristic (ROC) curve (AUC). Their cutoff values were determined using the maximum Youden’s Method. All statistical analyses were conducted using commercial software (SPSS, v25.0, Chicago, USA; Medcalc, v11.4, Mariakierke, Belgium). Statistical significance was considered for P < 0.05.