The authors have read the journal's policy and the authors of this manuscript have the following competing interests: CCC: Research grant Bayer HealthCare (solely materials of the study), speakers bureau Bayer HealthCare. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials.
Conceived and designed the experiments: CCC MFR KN. Performed the experiments: MJS CCC KN HH OD. Analyzed the data: MJS CCC KN HH OD. Contributed reagents/materials/analysis tools: MJS CCC KN HH MFR OD. Contributed to the writing of the manuscript: MJS CCC KN HH MFR OD.
To evaluate the use of diffusion-weighted MRI (DW-MRI) and volume measurements for early monitoring of antiangiogenic therapy in an experimental tumor model.
23 athymic nude rats, bearing human colon carcinoma xenografts (HT-29) were examined before and after 6 days of treatment with regorafenib (n = 12) or placebo (n = 11) in a clinical 3-Tesla MRI. For DW-MRI, a single-shot EPI sequence with 9 b-values (10–800 s/mm2) was used. The apparent diffusion coefficient (ADC) was calculated voxelwise and its median value over a region of interest, covering the entire tumor, was defined as the tumor ADC. Tumor volume was determined using T2-weighted images. ADC and volume changes between first and second measurement were evaluated as classifiers by a receiver-operator-characteristic (ROC) analysis individually and combined using Fisher's linear discriminant analysis (FLDA).
All ADCs and volumes are stated as median±standard deviation. Tumor ADC increased significantly in the therapy group (0.76±0.09×10−3 mm2/s to 0.90±0.12×10−3 mm2/s; p<0.001), with significantly higher changes of tumor ADC than in the control group (0.10±0.11×10−3 mm2/s vs. 0.03±0.09×10−3 mm2/s; p = 0.027). Tumor volume increased significantly in both groups (therapy: 347.8±449.1 to 405.3±823.6 mm3; p = 0.034; control: 219.7±79.5 to 443.7±141.5 mm3; p<0.001), however, the therapy group showed significantly reduced tumor growth (33.30±47.30% vs. 96.43±31.66%; p<0.001). Area under the curve and accuracy of the ADC-based ROC analysis were 0.773 and 78.3%; and for the volume change 0.886 and 82.6%. The FLDA approach yielded an AUC of 0.985 and an accuracy of 95.7%.
Regorafenib therapy significantly increased tumor ADC after 6 days of treatment and also significantly reduced tumor growth. However, ROC analyses using each parameter individually revealed a lack of accuracy in discriminating between therapy and control group. The combination of both parameters using FLDA substantially improved diagnostic accuracy, thus highlighting the potential of multi-parameter MRI as an imaging biomarker for non-invasive early tumor therapy monitoring.
Monitoring the response to anti-cancer treatment is an integral part of oncology. With the introduction of novel molecular cancer therapies to clinical routine it has become apparent that conventional, solely morphology-based imaging criteria, such as the Response Evaluation Criteria in Solid Tumors (RECIST)
DW-MRI is a method to visualize and quantify the mobility of water molecules in the observed tissue
The novel oral multi-kinase inhibitor regorafenib exhibits anti-angiogenic and anti-proliferative effects on glioblastoma, breast, and renal cell carcinoma xenografts
The purpose of this study was to evaluate quantitative DW-MRI and tumor growth measurements, individually and combined using a discriminant analysis approach, as means of distinguishing between therapy and control group of human colorectal carcinoma in rats under regorafenib or placebo therapy. We hypothesized that the combination of both approaches outperforms each classifier individually and can be used to monitor anti-angiogenic therapy non-invasively.
This study was approved by the Government of Upper Bavaria Committee for Animal Research (Gz.55.2-1-54-2532-33-10) and was carried out in accordance with the guidelines of the National Institute of Health for the care and use of laboratory animals. For the experiments twenty-three female athymic rats (7–8 weeks old, Harlan Laboratories Inc., Indianapolis, IN) were used. 2×106 cells of the human colon carcinoma cell line HT-29 (ATCC HTB-38) suspended in a total volume of 0.5 mL as a 1∶1 mixture of phosphate buffered saline pH 7.4 (PBS) and Matrigel (BD Biosciences, San Jose, CA) were injected subcutaneously into the left flanks. Prior to MRI the xenografts were allowed to grow to a reasonable size for imaging of approximately 400 mm3 (assessed by daily caliper measurements in three dimensions (
The therapy group was administered 10 mg/kg body weight of regorafenib daily, formulated as a solution in polypropylene glycol/PEG400/Pluronic F68 (42.5/42.5/15 + 20% Aqua), via gastric gavage, using a dedicated 16-gauge curved buttoned cannula. The control group received volume-equivalent applications of the regorafenib solvent daily.
Prior to MR imaging, animals were anaesthetized with isoflurane (5% for induction, 2.5% for maintenance, administered in pure oxygen). Scans were conducted on a clinical 3-Tesla whole-body MRI system (MAGNETOM Verio, Siemens Healthcare, Erlangen, Germany) with a small 4-channel flex coil (Siemens Healthcare, Erlangen, Germany).
DW-MRI was performed using a diffusion-weighted single-shot spin-echo sequence with echoplanar imaging (EPI) readout. A modified monopolar diffusion encoding scheme
Parameter | DW-MRI | T2-weighted MRI |
Acquisition plane | Axial | Axial |
Repetition time (ms) | 2500 | 9560 |
Echo time (ms) | 55 | 91 |
Signal averages | 8 | 3 |
Acquisition matrix | 68×52 | 192×192 |
Reconstructed matrix | 136×104 | 192×192 |
Field of view (mm2) | 65×50 | 60×60 |
Slice thickness (mm) | 2 | 1.5 |
Slice gap (mm) | 0.4 | 0 |
Number of slices | 12 | 35 |
Parallel imaging factor | 2 (GRAPPA) | 2 (GRAPPA) |
Fat supression | On | Off |
b-values (s/mm2) | 10, 25, 50, 80, 130, 200, 350, 550, 800 | - |
Acquisition time (min) | 10:08 | 6:53 |
Image analysis was performed on a dedicated workstation using our in-house software PMI (Platform for Research in Medical Imaging)
Prior to quantitative analysis, the diffusion-weighted images were rigidly registered along the b-value-dimension using a Fourier cross-correlation method to keep bulk-motion from affecting the diffusion coefficients. The ADC of each voxel was calculated by non-linear least-squares fitting of the measured signal intensities from all acquired b-values to the monoexponential diffusion model:
The ADC maps display a prominent increase, which is also reflected in the median tumor ADC value for this animal: ADCB = 0.762×10−3mm2/s at day 0, ADCF = 1.137×10−3mm2/s at day 7.
Obtaining quantitative parameters from MRI measurements is highly dependent on region of interest (ROI) placement, which often suffers from poor reproducibility. To obtain robust results and to minimize subjective influences on the ROI definition, we defined a 3D volume of interest (VOI) covering the entire tumor on multiple slices of the diffusion-weighted data for each animal and measurement (
Total ROI extends over 10 slices for this measurement.
The tumor volumes were determined based on the morphologic
Total ROI extends over 22 slices for this measurement.
All statistical analyses were performed using the statistical computing language R
Additionally, for each group the linear correlation between ΔADC and ΔVOL was determined using Pearson's product-moment correlation. To assess if the combination of both parameters, ΔADC and ΔVOL, increases the ability to distinguish between therapy and control group compared to the individual classifiers, Fisher's linear discriminant analysis
Median tumor ADCs with standard deviations and the results from the statistical analyses are summarized in
Although significantly different, ΔADC and ΔVOL display distinctive overlaps between the two groups. The result from FLDA demonstrates a marked improvement in the group discrimination with nearly no overlap, resulting in a highly significant difference.
The combined approach using FLDA outperforms the use of ΔADC and ΔVOL notably. Optimal sensitivity and specificity for each parameter and the corresponding thresholds are summarized in
Group | ADCB (10−3mm2/s) | ADCF (10−3mm2/s) | ΔADC (10−3mm2/s) |
Therapy | 0.76±0.09 |
0.90±0.12 |
+0.10±0.11 |
Control | 0.73±0.09 | 0.75±0.07 |
+0.03±0.09 |
Note: ADCB: baseline tumor ADC; ADCF: follow-up tumor ADC, ΔADC: tumor ADC changes between measurements.
*Therapy ADCB vs. therapy ADCF: p<0.001.
Therapy ADCF vs. control ADCF: p<0.001.
Therapy ΔADC vs. control ΔADC: p = 0.027
Classifier | AUC | Threshold | Sensitivity | Specificity | Accuracy |
ΔADC | 0.773 |
0.033×10−3mm2/s | 91.7% | 63.6% | 78.3% |
ΔVOL | 0.886 | +61.35% | 81.8% | 83.3% | 82.6% |
FLDA | 0.985 |
+0.139[a.u.] | 91.7% | 100% | 95.7% |
Note: ΔADC: tumor ADC changes between measurements, ΔVOL: tumor volume changes between measurements, FLDA: result from Fisher's linear discriminant analysis, AUC: area under the curve.
*AUC using ΔADC vs. AUC using FLDA: p = 0.035.
Median tumor volumes with standard deviations and the results from the statistical analyses are summarized in
Panel (a) illustrates thescatterplot of ΔVOL vs. ΔADC for each tumor. The solid grey line represents the optimal threshold determined by the ROC curve analysis; the linear regressions for each group (dashed line for therapy, dotted line for control) are annotated with Pearson's correlation coefficient r and p-value. (b) Results from the FLDA-derived linear combination of ΔADC and ΔVOL (FLDA = 0.0033×ΔVOL[%] - 1.0366×ΔADC[10−3 mm2/s]).
Group | VOLB (mm3) | VOLF (mm3) | ΔVOL (%) |
Therapy | 347.8±449.1 |
405.3±823.6 |
33.30±47.30 |
Control | 219.7±79.5 |
443.7±141.5 |
96.43±31.66 |
Note: VOLB: baseline tumor volume; VOLF: follow-up tumor volume, ΔVOL: tumor volume changes between measurements.
*Therapy VOLB vs. therapy VOLF: p = 0.034.
Control VOLB vs. control VOLF: p<0.001.
Therapy ΔVOL vs. control ΔVOL: p<0.001.
The results from FLDA are also shown in
In this study, we used MRI in an experimental colon carcinoma model to evaluate the influence of the recently FDA-approved multi-kinase inhibitor regorafenib
Diffusion-weighted MRI revealed microstructural changes in the tumorous tissue reflected by an increased water diffusivity induced by a one-week therapy with regorafenib. The significant increase in tumor ADC in the therapy group is likely due to apoptosis, which was shown to be significantly unregulated by regorafenib therapy in the same tumor cell line (HT29)
The significant increase in tumor ADCs in the therapy group and more importantly the significant differences in the observed tumor ADC changes between the two groups mark DW-MRI as a potential biomarker for monitoring of molecular cancer therapy, including multi-tyrosine kinase inhibitors such as regorafenib. While Thoeny et al. have reported an initial decrease in water diffusivity in the first hours after anti-cancer therapy initiation
While regorafenib seems to have a significant effect on water diffusivity, the overlap in the observed tumor ADC changes between therapy and control group leads to a less accurate discrimination and therefore to a limitation of the therapy monitoring capabilities of DW-MRI. To increase diagnostic accuracy it is therefore advisable to combine ADC measurements with additional information about tumor progression, such as tumor volume.
The tumor volume increased significantly between baseline and follow-up measurement in both groups. This strongly indicates that the Response Evaluation Criteria in Solid Tumors (RECIST)
Fisher's linear discriminant analysis is an effective method to combine two or more classifiers in separation problems
The linear combination of ADC and volume changes calculated with FLDA and also the thresholds determined with the ROC curve analyses are specific to the tumor type, the therapy and the time interval between measurements. These parameters may have to be reevaluated according to the respective settings. However, the concept of the method presented in this study to integrate morphological and functional information as complementing parameters should remain valid.
For further analysis, it may be possible to gain additional insight on the tumor physiology by investigating the histogram shape (e.g. variance or skewness) of the ADC distribution inside the VOI (
The median tumor ADC increased in both cases (therapy: 0.72×10−3 mm2/s to 0.91×10−3 mm2/s, control: 0.73×10−3 mm2/s to 0.82×10−3 mm2/s), however, the therapy tumor grew by 36%, while the control group tumor grew by 76%.
Using quantitative DW-MRI, we found that therapy of human colon carcinoma xenografts with the multi-tyrosine kinase inhibitor regorafenib significantly increased water diffusivity in tumorous tissue after 6 days of treatment. We also observed that regorafenib significantly reduced tumor growth compared to the control group. Using either tumor ADC changes or tumor growth to distinguish between therapy and control group resulted in diagnostic accuracy of about 78% and 83%, respectively, which we consider not sufficient for an imaging biomarker. The approach to combine both parameters using Fisher's linear discriminant analysis, substantially improved the accuracy to about 96%, thus highlighting the potential of multi-parameter MRI as an imaging biomarker for non-invasive early tumor therapy monitoring.
(XLSX)
(XLSX)
(XLSX)