Processing MRI data for analysis of tumors

Abstract

Dynamic contrast enhanced MRI data and dynamic susceptibility contrast MRI data for a volume of the patient's body including a tumor are used in order to identify the risk of lymph node metastasis. A volume of interest enclosing the tumor is identified and the MRI data for the volume of interest is processed in order to identify one or more parameters relating to the transverse relaxation rate and/or to dynamic changes in the transverse relaxation rate. A comparison is made to previously obtained similar parameters for other patients having similar tumors that are either known to exhibit lymph node metastasis or are known not to exhibit lymph node metastasis in order to determine if the MRI data indicates that the patient has a tumor at greater risk of exhibiting lymph node metastasis or a tumor with a lesser risk of exhibiting lymph node metastasis.

Claims

1. A method for processing MRI data for analysis of rectal tumors, the method comprising: obtaining dynamic contrast enhanced MRI data and dynamic susceptibility contrast MRI data for a volume of the patient's body including a rectal tumor; identifying a volume of interest enclosing the rectal tumor; processing the MRI data for the volume of interest in order to identify one or more parameters relating to the transverse relaxation rate and/or to dynamic changes in the transverse relaxation rate, wherein the one or more parameters relating to the transverse relaxation rate and/or to dynamic changes in the transverse relaxation rate include one or more of: the peak change in tumor R.sub.2*, the dynamic peak change in tumor R.sub.2*, a maximum of the peak change or the dynamic peak change, and/or an integral for dynamic peak change in tumor R.sub.2* over a time period; comparing the one or more parameters to previously obtained similar parameters for multiple other patients having similar rectal tumors that are either known to exhibit lymph node metastasis or are known not to exhibit lymph node metastasis; and thereby determining if the MRI data indicates that the rectal tumor is at a greater risk of exhibiting lymph node metastasis or at a lesser risk of exhibiting lymph node metastasis.

2. The method as claimed in claim 1, comprising determining if the MRI data indicates a patient with a rectal tumor at greater risk of exhibiting lymph node metastasis by comparison of the one or more parameters with the previously obtained similar parameters and determining if the one or more parameters are similar to previously obtained similar parameters for patients having similar rectal tumors that are known to exhibit lymph node metastasis.

3. The method as claimed in claim 1, comprising determining if the MRI data indicates a patient with a rectal tumor at greater risk of exhibiting lymph node metastasis by comparison of the one or more parameters with the previously obtained similar parameters and determining if the one or more parameters are lower than previously obtained similar parameters for patients having similar rectal tumors that are known not to exhibit lymph node metastasis.

4. The method as claimed in claim 1, wherein the comparison includes the use of a threshold value determined based on the previously obtained similar parameters.

5. The method as claimed in claim 4, wherein the threshold value is a value lying between the previously obtained similar parameters for patients having similar rectal tumors that are known to exhibit lymph node metastasis and the previously obtained similar parameters for patients having similar rectal tumors that are known not to exhibit lymph node metastasis.

6. The method as claimed in claim 1, wherein the tumor R.sub.2* is wash-in R.sub.2* peak enhancement.

7. The method as claimed in claim 1, wherein both of the dynamic contrast enhanced MRI data and the dynamic susceptibility contrast MRI data are obtained in a single dynamic examination.

8. A system comprising dynamic contrast enhanced MRI data and dynamic susceptibility contrast MRI data for a volume of the patient's body including a rectal tumor along with data processing apparatus for processing the MRI data for analysis of rectal tumors, wherein the data processing apparatus comprises: an MRI data receiving module for receiving the dynamic contrast enhanced MRI data and dynamic susceptibility contrast MRI data; a volume of interest identification module for generating and/or receiving data identifying a volume of interest enclosing the rectal tumor; data storage; and an MRI data processing module for processing the MRI data for the volume of interest and being arranged to: identify one or more parameters relating to the transverse relaxation rate and/or to dynamic changes in the transverse relaxation rate, wherein the one or more parameters relating to the transverse relaxation rate and/or to dynamic changes in the transverse relaxation rate include one or more of: the peak change in tumor R.sub.2*, the dynamic peak change in tumor R.sub.2*, a maximum of the peak change or the dynamic peak change, and/or an integral for dynamic peak change in tumor R.sub.2* over a time period, compare the one or more parameters to previously obtained similar parameters for multiple other patients having similar rectal tumors that are either known to exhibit lymph node metastasis or are known not to exhibit lymph node metastasis, wherein the previously obtained similar parameters for rectal tumors are stored on the data storage and accessed therefrom by the MRI data processing module, and thereby determine if the MRI data indicates that the rectal tumor is at a greater risk of exhibiting lymph node metastasis or at a lesser risk of exhibiting lymph node metastasis.

9. The data processing apparatus as claimed in claim 8, wherein the data processing module is arranged to determine if the MRI data indicates a patient with a rectal tumor at greater risk of exhibiting lymph node metastasis by comparison of the one or more parameters with the previously obtained similar parameters and determining if the one or more parameters are similar to previously obtained similar parameters for patients having similar rectal tumors that are known to exhibit lymph node metastasis.

10. The data processing apparatus as claimed in claim 8, wherein the data processing module is arranged to determine if the MRI data indicates a patient with a rectal tumor at greater risk of exhibiting lymph node metastasis by comparison of the one or more parameters with the previously obtained similar parameters and determining if the one or more parameters are lower than previously obtained similar parameters for patients having similar rectal tumors that are known not to exhibit lymph node metastasis.

11. The data processing apparatus as claimed in claim 8, wherein the data processing apparatus is arranged to compare the one or more parameters to previously obtained similar parameters using a threshold value determined based on the previously obtained similar parameters, wherein the threshold value is a value lying between the previously obtained similar parameters for patients having similar rectal tumors that are known to exhibit lymph node metastasis and the previously obtained similar parameters for patients having similar rectal tumors that are known not to exhibit lymph node metastasis.

12. The data processing apparatus as claimed in claim 8, wherein the tumor R.sub.2* is wash-in R.sub.2* peak enhancement.

13. The data processing apparatus as claimed in claim 8, comprising an MRI control module for controlling an MRI apparatus in order to obtain both of the dynamic contrast enhanced MRI data and the dynamic susceptibility contrast MRI in a single dynamic examination.

14. A non-transitory computer readable medium comprising a computer program product containing instruction that, when executed, will configure a data processing apparatus to: receive dynamic contrast enhanced MRI data and dynamic susceptibility contrast MRI data for a volume of the patient's body including a rectal tumor; receive data identifying a volume of interest enclosing the rectal tumor; process the MRI data for the volume of interest in order to identify one or more parameters relating to the transverse relaxation rate and/or to dynamic changes in the transverse relaxation rate, wherein the one or more parameters relating to the transverse relaxation rate and/or to dynamic changes in the transverse relaxation rate include one or more of: the peak change in tumor R.sub.2*, the dynamic peak change in tumor R.sub.2*, a maximum of the peak change or the dynamic peak change, and/or an integral for dynamic peak change in tumor R.sub.2* over a time period; compare the one or more parameters to previously obtained similar parameters for multiple other patients having similar rectal tumors that are either known to exhibit lymph node metastasis or are known not to exhibit lymph node metastasis; and thereby determine if the MRI data indicates that the rectal tumor is at a greater risk of exhibiting lymph node metastasis or at a lesser risk of exhibiting lymph node metastasis.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Certain preferred embodiments will now be described by way of example only and with reference to the accompanying drawings in which:

(2) FIGS. 1(a) and 1(b) are box and whisker plots showing the parameter R.sub.2*-peak.sub.enh when comparing patients with and without nodal metastasis based on volumes of interest (VOI) by (a) Reader 1 and (b) Reader 2;

(3) FIGS. 2(a) and 2(b) are plots of high temporal resolution ΔR.sub.1-(a) and ΔR.sub.2* (b) time-curves for a 66-year-old male histologically diagnosed with adenocarcinoma;

(4) FIG. 3 is a plot of high temporal resolution ΔR.sub.2* time-curves from four rectal cancer patients;

(5) FIG. 4 shows AIF automatically detected together with the average ΔR.sub.2* time-curves and corresponding gamma-variate-fitted curve during bolus first pass for a 67-year-old male histologically diagnosed with adenocarcinoma; and

(6) FIGS. 5(a) to 5(d) show selected K.sup.trans and k.sub.ep versus estimated K.sup.trans and k.sub.ep using a continuous dynamic acquisition (a and c) and a split dynamic acquisition (b and d), respectively, with the lower figures showing the corresponding residuals and residual norms obtained from the linear regression.

(7) FIG. 6 is a schematic illustration of a data processing apparatus for processing MRI data for analysis of tumors.

DETAILED DESCRIPTION OF THE INVENTION

(8) Introduction

(9) Current international guidelines recommend magnetic resonance imaging (MRI) as part of primary staging of rectal cancer. MRI helps identify patients with increased risk of local recurrence and the need of neoadjuvant chemoradiotherapy. Despite improvements in multimodal rectal cancer management during the past decades, with local recurrence rates below 10%, a considerable number of patients experience poor disease outcome resulting from metastatic disease progression. Reliable detection of metastatic lymph nodes (N+ stage), a main prognostic indicator of distant metastasis, is still a challenge in rectal cancer MRI. At present, the most accurate criterion is the morphological finding of irregular-contoured nodes with heterogeneous signal pattern. However, these MRI findings are associated with substantial misinterpretation and the diagnostic accuracy is relatively poor. A recent meta-analysis of 21 clinical studies showed an accuracy for lymph node staging of 71%.

(10) To improve the ability of MRI to predict patients at risk of developing metastatic disease, considerable interest is focusing on the tumor microenvironment. Dynamic contrast-based MRI is an increasingly popular method for tumor characterization, enabling quantitative assessment of phenotypic properties. The most commonly used dynamic acquisition is dynamic contrast-enhanced (DCE) MRI, enabling assessment of tissue properties such as capillary permeability and extracellular volume fraction. There is currently no consensus concerning the role of DCE-MRI in rectal cancer patients.

(11) Dynamic susceptibility contrast (DSC) MRI is most commonly used to measure perfusion in the brain. However, several studies have also shown the feasibility of using this approach to assess phenotypic characteristics of breast tumors. To our knowledge, DSC-MRI analysis has never before been applied in rectal cancer. Recognizing that rectal tumors are highly angiogenic, it is hypothesized that DSC-MRI may provide valuable information for assessing pathophysiological properties in this patient group.

(12) Multiple echoes can be acquired in a high temporal resolution dynamic contrast-based MRI sequence, allowing quantitative measurements of the dynamic change in both longitudinal- (R.sub.1=1/T.sub.1) and transverse relaxation rate (R.sub.2*=1/T.sub.2*). As a result, DCE- and DSC-data can be obtained during a single dynamic examination, thus yielding information on pathophysiological properties related to both tissue-permeability and perfusion. In earlier studies, the feasibility of a multi-echo MRI technique was demonstrated through both simulations and in the assessment of breast cancer.

(13) The aim of this study was to implement a dynamic contrast-based multi-echo MRI sequence in the assessment of rectal cancer, and to evaluate associations between clinicopathological data and the attainable DCE- and DSC-MRI parameters.

(14) Materials and Methods

(15) Study Patients

(16) The investigation was performed within the prospective biomarker study OxyTarget—Functional MRI of Hypoxia-Mediated Rectal Cancer Aggressiveness (NCT01816607), approved by the Institutional Review Board and the Regional Committee for Medical and Health Research Ethics of South East Norway. The study was performed in accordance with the Helsinki declaration and written informed consent for participation.

(17) The principal inclusion criterion was histologically confirmed rectal cancer scheduled to radical treatment. The patient cohort within the present study was enrolled between March 2014 and June 2015 and all cases had primary pelvic surgery. The resected tumor specimens underwent standard histopathologic staging (pTN), including determination of the absence or presence of extramural venous invasion. Patient and tumor characteristics are provided in Table 1 below.

(18) TABLE-US-00001 TABLE 1 Patient and tumor characteristics No. of patients 17 Gender Male 11 (64.7%) Female 6 (35.3%) Median age (years).sup.a 66 (50-88) Male 66 (52-88) Female 71 (50-77) Median tumor volume (cm.sup.3).sup.a,b 18.8 (4.5-64.0) rTNM stages.sup.c mrT1 2 (11.8%) mrT2 5 (29.4%) mrT3 8 (47.1%) mrT4 2 (11.8%) mrN0 10 (58.8%) mrN1 6 (35.3%) mrN2 1 (5.9%) rM0 16 (94.1%) rM1 1 (5.9%) Surgery Low anterior resection 13 (76.5%) Abdominal perineal resection 3 (17.6%) Transanal endoscopic microsurgery 1 (5.9%) pTN stages.sup.d pT1 4 (23.5%) pT2 4 (23.5%) pT3 8 (47.1%) pT4 1 (5.9%) pN0 10 (58.8%) pN1 6 (35.3%) pN2 1 (5.9%) Extramural venous invasion.sup.d Yes 6 (35.3%) No 11 (64.7%) NOTE. Except where indicated, data are numbers of patients, with percentages in parentheses .sup.aNumbers in parentheses are age ranges .sup.bMean tumor volumes calculated from two independent radiologists' tumor delineations in magnetic resonance images (MRI) .sup.cAssessed with MRI or computed tomography (CT) according to the tumor-node-metastasis system. Prefix ‘mr’ denotes MRI-assessed staging, prefix ‘r’ denotes radiologically assessed staging (MR and/or CT)

(19) MRI Acquisition

(20) In this pilot study, 24 patients were examined, of which five were excluded because of bowel motion and two because of image artifacts. Imaging was performed on a Philips Achieva 1.5T system (Philips Healthcare, Best, The Netherlands) with NOVA Dual HP gradients (33 mT/m maximum gradient amplitude, 180 T/m/s slew rate). A five channel cardiac coil with parallel imaging capabilities was applied. Glucagon (1 mg/ml, 1 ml intramuscularly) and buscopan (20 mg/ml, 1 ml intravenously) were given to reduce bowel peristalsis. Conventional high-resolution fast spin-echo T.sub.2-weighted images of the pelvic cavity and rectum were obtained in the sagittal and transversal planes as well as perpendicular to the tumor axis (TR=2820-3040 ms, TE=80 ms, acquisition matrix=256×230, slice thickness=2.5 mm, number of excitations=6 and echo train length=20).

(21) Dynamic contrast-based multi-echo data were acquired by a 3D T.sub.1-weighted multi-shot EPI sequence with three echoes. TR=39 ms, TE1=4.6 ms, echo spacing=9.3 ms, flip angle=28°, EPI factor=9. The acquired matrix size was 92×90 over a 180×180 mm field-of-view (FOV). Twelve slices were acquired with a thickness of 10 mm. The temporal resolution was approximately 2 s/imaging volume with 60 dynamic series acquired. A ProSet fat suppression technique was applied along with a parallel imaging (SENSE) factor of 1.7 in the RL direction. All slices were acquired parallel to the T.sub.2-weighted images perpendicular to the tumor axis.

(22) The dynamic multi-echo sequence was interleaved with a high spatial resolution 3D T.sub.1-weighted TFE sequence (THRIVE), as part of a split-dynamic MRI framework detailed in a recent study (13). In this framework, one set of THRIVE baseline images was initially acquired. The dynamic multi-echo sequence was then started and 5-7 baseline acquisitions were obtained prior to CA administration. The multi-echo acquisition was repeated approximately 30 times during the wash-in phase, immediately followed by the first post-contrast THRIVE acquisition. During the intermediate and late post-contrast phase, 6 split sessions were conducted, acquiring 4 multi-echo data sets and interleaved THRIVE images. In addition, a series of 14 multi-echo data sets were acquired after the last THRIVE segment to improve curve fitting for kinetic modeling. To avoid non-steady state effects, a 2500 ms dummy repetition (i.e., no data sampling) was run in the beginning of each multi-echo segments. Only results from the dynamic multi-echo data are reported in this work.

(23) A dose of 0.2 mL/kg body weight of gadolinium-based CA (Dotarem® 279.3 mg/mL, Guerbet, Roissy, France) was injected as a bolus (3 mL/s) directly followed by 20 ml of physiologic saline solution.

(24) Simulations

(25) Given that the multi-echo acquisition was run as part of a split-dynamic MRI framework, simulations were performed to investigate whether the splitting of dynamic time-series influences the reliability of parameter estimation in comparison with a continuous acquisition. The simulations were a series of Monte Carlo simulations to investigate whether a split dynamic acquisition technique influences the reliability of parameter estimation in dynamic MR-measurements in comparison with a conventional continuous acquisition. The simulations were performed using in-house algorithm developed in MATLAB (The MathWork Inc. version 7.14.0.739 (R2012a), Natick, Mass., USA).

(26) A system input function was simulated as a gamma variate function with an additional exponential term in order to simulate steady state effects. Corresponding tissue CA-concentration time curves were simulated according to the extended Tofts model (7) by randomly selecting the kinetic parameters within a defined range of values; 0.13-3.17 min.sup.−1 for K.sup.trans, 0-80% for v.sub.e and 0-20% for v.sub.p. The selected K.sup.trans- and v.sub.e-range were based on values previously observed in a clinical rectal cancer study, and was defined as the mean value±two standard deviations (SD). The bolus-arrival-time (BAT) was randomly selected between 0-10 seconds, and a random Gaussian noise corresponding to a signal-to-noise ratio (SNR) of 20 at a temporal resolution of 2 seconds was added to the CA-concentration time curves. The selected SNR was based on values observed in the patient data.

(27) To investigate the reliability of the split dynamic acquisition technique in comparison with a conventional continuous acquisition, one thousand simulations were performed with a fixed temporal resolution of 2 seconds. The system input function and tissue response curves were then resampled using the split dynamic scheme as implemented in the split dynamic technique, before adding noise based on the selected SNR. Kinetic parameter estimates for the continuous- and splitted CA concentration time curves were then obtained.

(28) The goodness of fit for a given parameter estimation was evaluated by plotting the nominal parameter value against the estimated value and performing linear regression analysis, including model residuals and residual norms to the resulting plots. Mann-Whitney U test was used to test the difference between the split dynamic acquisition and the continuous acquisition, with the null hypothesis that the data obtained from the two acquisition methods are samples from continuous distributions with equal medians. A statistical significance level of 5% was used.

(29) Image Analysis

(30) Diagnostic radiological TNM stages were assessed according to international guidelines and the 7.sup.th edition TNM staging system. Guided by T.sub.2-weighted and diffusion-weighted images, two radiologists with 14 and 7 years of experience (Reader 1 and Reader 2) independently contoured whole-tumor volume-of-interests (VOI) by means of free-hand delineations.

(31) Image post-processing was performed using the nordicICE software (NordicNeuroLab, Bergen, Norway). CA concentration-time curves were estimated from the first echo using the spoiled gradient echo (SPGR) signal equation, denoted S.sub.SPGR(t), with the T.sub.2 term ignored and assuming a linear relationship between the CA concentration and change in R.sub.1 (1/T.sub.1):

(32) $\begin{array}{cc}{C}_t\left(t\right)=\frac{\left(1/T_1\left(t\right)\right)-1/T_{1,0}}{r_1}& 1]\end{array}$

(33) where r.sub.1 is the longitudinal relaxivity of the CA and T.sub.1,0 is the T.sub.1 of the tissue in the absent of CA. The signal from a SPGR is given by:

(34) $\begin{array}{cc}{S}_{\mathrm{SPGR}}\left(t\right)=S_0\frac{\sin \alpha \left(1-e^{-\mathrm{TR}/T_1\left(t\right)}\right)}{1-\cos \alpha \left(e^{-\mathrm{TR}/T_1\left(t\right)}\right)}& 2]\end{array}$

(35) where S.sub.0 is proportional to the thermal equilibrium magnetization, TR is the repetition time of the sequence, and α is the flip angle. Pre-contrast T.sub.1-values were estimated in six patients using a modified Look-Locker inversion recovery (MOLLI) technique. Average T.sub.1-values (±standard deviation) were measured to 1528±40 ms in blood and 1354±103 ms in cancer tissue and used as a fixed T.sub.1-baseline for SPGR signal conversion in arteries and tumors, respectively. The measured T.sub.1-values in blood agreed well with literature values.

(36) The CA concentration-time curves were then analyzed on a voxel-by-voxel basis using the extended Tofts pharmacokinetic model:

(37) $\begin{array}{cc}{C}_t\left(t\right)=K^{\mathrm{trans}}{C}_a\left(t\right).Math.e^{-\frac{K^{\mathrm{trans}}}{v_e}t}+v_p{C}_a\left(t\right)& 3]\end{array}$

(38) where .Math. represents the convolution operator and K.sup.trans is the volume transfer constant between the plasma volume, v.sub.p, and the extravascular extracellular space volume, v.sub.e. An arterial input function (AIF), C.sub.a, was obtained for each patient by extracting the CA concentration-time curve from an artery supplying the region of interest (ROI) using an automatic cluster-algorithm, and an average AIF was generated and used for quantitative analysis. The enhancement delay between the AIF- and the VOI-signal were manually measured and included as a fixed variable in the kinetic model. Voxel-wise CA concentration-time curves were also analyzed by estimating the maximum peak enhancement (Peak.sub.enh), time-to-peak (TTP), area under the curve (AUC), wash-in and wash-out rate.

(39) From the multi-echo data, voxel-wise dynamic change in R.sub.2* was estimated by assuming a mono-exponential signal-dependent change as a function of TE:
SI(t.sub.m,TE.sub.n)=SI.sub.0(t.sub.m)e.sup.−TE.sup.n*.sup.R.sup.2.sup.*(t.sup.m.sup.)  4]
where the subscript m=1, . . . , M labels the repetition of the multi-echo acquisition, n=1, . . . , N labels the echo number, and SI.sub.0(t.sub.m) is the initial signal amplitude. The TE-dependent signal was then fitted to Eq. 4 using a standard least squares fitting algorithm to estimate voxel-wise R.sub.2* values. Dynamic ΔR.sub.2* data was further analyzed using the established tracer kinetic model for DSC-MRI (19), expressing the relationship between the tissue response and the AIF, yielding semi-quantitative analysis of blood flow (BF) and blood volume (BV). Corresponding mean transit time (MTT) is given by the central volume principle: MTT=BV/BF. To minimize contamination due to CA recirculation and leakage, the dynamic ΔR.sub.2* data was fitted to a gamma-variate function. For each patient, an AIF was obtained by extracting the ΔR.sub.2*-time curve from an artery supplying the ROI using an automatic cluster-algorithm. The voxel-wise ΔR.sub.2*-time curves were also analyzed by estimating the maximum dynamic peak change (R.sub.2-peak.sub.enh) and area under the curve (R.sub.2*-AUC). Due to the early onset of the first intravascular CA passage, only data from the first multi-echo segment was used for DSC analysis.

(40) Statistical Analysis

(41) Statistical analysis was performed using R version 2.10.1 (R Foundation for Statistical Computing, Vienna, Austria) and MATLAB R2015a version 8.5.0 (Mathworks, Inc., Natick, Mass., USA). The following pathological stages were grouped in the statistical analysis; T1 and T2, T3 and T4, and N1 and N2. A Mann-Whitney U test was used to evaluate associations between kinetic parameters and clinicopathologic data. A p-value<0.05 was considered significant. Statistically significant parameters were further evaluated using receiver operator characteristic (ROC) curve statistics. Intraclass correlation coefficient (ICC) for continuous variables was used to evaluate interobserver agreement for the measured whole-tumor VOI.

(42) Results

(43) MRI Analysis

(44) Table 2 summarizes the associations between kinetic parameters attainable with the dynamic contrast-based multi-echo sequence and histopathologic evaluation of the surgically resected specimens. For patients with histologically confirmed nodal metastasis, the primary tumor showed a significantly lower R.sub.2*-peak.sub.enh than patients without nodal metastasis (See FIG. 1), corresponding to a p-value of 0.005 for Reader 1 and 0.019 for Reader 2, and an area under the ROC-curves (sensitivity, specificity) of 90 (86%, 90%) and 84 (86%, 80%), respectively. The box and whisker plots of FIGS. 1(a) and 1(b) illustrate the median (via the line inside the box) and the mean (via the circle inside box) values, interquartile range (the box itself), as well as minimal and maximal values (via the whisker lines extending above and below the box). For Reader 1, T1 and T2 cases showed a significantly shorter TTP (p=0.046) and higher BV (p=0.021), compared to T3 and T4 cases. The corresponding area under the ROC-curves (sensitivity, specificity) was 79 (71%, 90%) and 83 (71%, 90%), respectively. For Reader 2, no significant associations were found between the DCE parameters and the pT-stage. A mismatch between mrN and pN was found in 6 of 17 patients (35.3%), corresponding to an accuracy of 64.7%.

(45) TABLE-US-00002 TABLE 2 Comparison of the functional DCE- and DSC-MRI parameters with clinicopathologic data and corresponding p-values. The table shows mean parametric values based on whole-tumor medians extracted using VOI from reader 1 (A) and reader 2 (B). A: Reader 1 P-values Pathologic T-stage T1/T2 vs Pathologic N-stage P-values Parameters T1 T2 T3 T4 T3/T4 N− N+ N− vs N+ DCE-MRI analysis K.sup.trans 0.142 0.188 0.182 0.171 0.673 0.183 0.171 0.475 k.sub.ep 0.562 0.631 0.559 0.524 0.606 0.631 0.524 0.364 v.sub.e 18.3 25.2 30.4 30.5 0.370 28.2 30.5 0.962 v.sub.p 6.20 7.33 5.09 4.67 0.059 6.82 4.70 0.161 AUC 242.4 360.1 387.7 362.6 0.321 370.3 278.9 0.315 Peak.sub.enh 1.36 1.90 1.92 1.70 0.541 1.90 1.70 0.417 TTP 37.1 34.7 110.8 122.6 0.046 44.8 118.0 0.133 Wash In 0.061 0.106 0.071 0.053 0.370 0.078 0.063 0.315 Wash Out 0.0050 0.0055 0.0030 0.0016 0.093 0.0052 0.0016 0.161 DSC-MRI analysis BF 177.6 255.7 183.4 122.0 0.139 237.1 166.9 0.088 BV 22.1 21.1 15.9 12.4 0.021 20.6 14.6 0.109 MTT 6.60 4.83 4.54 4.90 0.200 4.67 4.90 0.887 R.sub.2*-AUC 445.7 520.4 431.0 310.8 0.167 498.2 339.9 0.070 R.sub.2*-Peak.sub.enh 25.5 39.9 29.9 15.3 0.321 35.4 23.4 0.005 B: Reader 2 P-values Pathologic T-stage T1/T2 vs Pathologic N-stage P-values Parameters T1 T2 T3 T4 T3/T4 N− N+ N− vs N+ DCE-MRI analysis K.sup.trans 0.150 0.192 0.180 0.169 1.000 0.187 0.169 0.364 k.sub.ep 0.591 0.541 0.580 0.469 0.815 0.562 0.469 0.601 v.sub.e 22.3 27.7 30.8 33.8 0.370 28.7 31.1 0.813 v.sub.p 5.0 7.4 5.2 4.7 0.423 6.4 4.7 0.601 AUC 230.2 363.6 380.5 375.7 0.236 371.7 278.9 0.475 Peak.sub.enh 1.32 1.84 1.91 1.76 0.481 1.88 1.71 0.417 TTP 40.2 75.7 113.6 194.3 0.321 107.2 118.0 0.475 Wash In 0.064 0.094 0.071 0.055 0.541 0.075 0.062 0.230 Wash Out 0.0062 0.0042 0.0018 −0.0006 0.321 0.0047 0.0011 0.230 DSC-MRI analysis BF 188.4 254.0 185.0 122.0 0.423 220.1 161.8 0.193 BV 20.7 18.6 15.8 12.4 0.167 18.9 14.1 0.230 MTT 7.04 4.63 4.56 5.08 0.321 4.70 5.08 0.887 R.sub.2*-AUC 413.9 512.3 426.4 298.8 0.321 502.7 315.0 0.070 R.sub.2*-Peak.sub.enh 23.4 39.8 29.6 14.6 0.423 36.0 23.0 0.019 T1: n = 4; T2: n = 4; T3: n = 8; T4: n = 1; N−: n = 10; N+: n = 7. P-values at the univariate analysis were obtained by Mann-Whitney U-test. Parameters showing statistically significant differences are highlighted in bold.

(46) FIG. 2 shows DCE- and DSC-data obtained from the multi-echo sequence in a sample case. The curves were extracted using a circular region of interest in the center of the tumor. Resulting parametric maps (no shown) representing K.sup.trans, k.sub.ep, BV and R.sub.2*-peak.sub.enh were also obtained. The tumor showed a fast wash-in and wash-out rate in the R.sub.1-data, and a strong transient increase during bolus first pass in the R.sub.2*-data.

(47) FIG. 3 shows ΔR.sub.2* time-curves from four cases. This figure demonstrates the added value of ΔR.sub.2* analysis of the primary tumor in distinguishing patients with and without nodal metastasis. Corresponding T.sub.2-weighted images merged with parametrical maps representing ΔR.sub.2-peak.sub.enh (not shown) were also obtained. The ΔR.sub.2* time-curves were extracted using a region of interest, drawn by an experienced radiologist and delineating the tumor at a central slice. Cases 1 and 2 showed a negative N-stage (N−), whereas cases 3 and 4 showed a positive N-stage (N+). Note the difference in ΔR.sub.2* time-curves where the N− tumors showed a strong transient increase during the bolus first-pass while the N+ tumors showed lower R.sub.2*-enhancement. This can also be seen in the parametric maps where the N+ tumors showed an overall low dynamic change in R.sub.2 whereas the N− tumors appear highly heterogeneous with hot-spots demonstrating a strong increase in R.sub.2*.

(48) FIG. 4 shows the measured AIF- and VOI-curves together with the corresponding gamma-variate fitted curve for a selected case. The patient is a 67-year-old male histologically diagnosed with adenocarcinoma. The figure shows the AIF automatically detected together with the average ΔR.sub.2* time-curves during bolus first pass. The corresponding gamma-variate-fitted curve is shown as dotted line. The average ΔR.sub.2* time-curves were extracted from a central slice. Parametric maps representing BV and R.sub.2*-peak.sub.enh (not shown) were also obtained as overlay on T.sub.2-weighted image. FIG. 4 demonstrates the importance of CA administration timing and acquisition length of the first multi-echo acquisition segment in order to include the first pass bolus.

(49) The ICC (95% CI) between the two readers was 0.95 (0.87-0.98) for all whole-tumor VOIs, 0.81 (0.32-0.96) for T.sub.1- and T2-stages tumor VOIs, and 0.95 (0.80-0.99) for T3- and T4-stages tumor VOIs.

(50) FIG. 5 shows the accuracy of K.sup.trans and k.sub.ep estimations from a continuous and split dynamic acquisition for a defined range of initially selected parametric values. K.sup.trans and k.sub.ep versus estimated K.sup.trans and k.sub.ep are shown with data using a continuous dynamic acquisition (a and c) and a split dynamic acquisition (b and d), respectively. The bottom figures show the corresponding residuals and residual norms obtained from the linear regression. The residuals indicated that the dispersion of data, and thus the uncertainty of parameter estimates, increases with higher parameter values for both K.sup.trans and k.sub.ep. However, the residuals were symmetrically distributed around zero in all cases suggesting absence of a systematic error in the parameter estimates. A Mann-Whitney U test showed that there was no significant difference in the estimation of neither K.sup.trans (p=0.71), k.sub.ep (p=0.81), v.sub.e (p=0.99) nor v.sub.p (p=0.71) using the two acquisition alternatives.

(51) A schematic illustration of a data processing apparatus 600 for processing MRI data for analysis of tumors is shown in FIG. 6. The data processing apparatus 600 includes an MRI data receiving module 601 for receiving dynamic contrast enhanced MRI data and dynamic susceptibility contrast MRI data for a volume of the patient's body including a tumor; a volume of interest identification module 602 for generating and/or receiving data identifying a volume of interest enclosing the tumor; an MRI data processing module 603 for processing the MRI data for the volume of interest; and an MRI control module 604 for controlling an MRI apparatus 605.

(52) Discussion

(53) The study described herein identified a significant association between R.sub.2*-peak.sub.enh in the primary rectal tumor by DSC-MRI and the nodal status evaluated by histopathology of the surgical specimen, showing an area under the ROC-curve of 90% and 84% for Reader 1 and 2, respectively. In comparison, radiological assessments of nodal status agreed with histopathological evaluation in 64.7% (11 of 17 patients). This finding is comparable to that of a recent meta-analysis (5). A lower value of R.sub.2*-peak.sub.enh was significantly associated with the presence of lymph node metastasis.

(54) One hypothesis is that the peak change in R.sub.2* reflects tissue blood supply, and that a low R.sub.2*-peak.sub.enh areas may be associated with hypoxic tissue caused by insufficient blood flow. The nodal status represents a main prognostic marker for metastatic progression and unfavorable survival. A positive N-stage will also have implications for treatment, as these patients should be considered for neoadjuvant therapy, and extended pelvic surgery to include all lymph nodes. DSC-MRI of rectal cancer can provide an important indicator for lymph node status, which will improve mrN staging prior to commencement of therapy.

(55) Given that the multi-echo acquisition was run as part of a split-dynamic MRI framework, simulations were performed to investigate whether the splitting of dynamic time-series may influence the reliability of parameter estimation in comparison with a continuous acquisition. According to the simulations, the split dynamic acquisition did not significantly affect the accuracy of kinetic parameter estimates compared with using a continuous acquisition. Uncertainties in the estimates were found to increase with increasing parametric value for both acquisition methods, with a somewhat larger error in for the split dynamic approach. This may indicate that the splitting process reduces the sensitivity to accurately estimate extremely abnormal kinetic values. However, this difference was not statistically significant, and the findings suggest that essential information is not lost in the splitting process, and that a split dynamic approach will provide sufficient information of the contrast enhancement kinetics for clinically relevant parametric values.

(56) Whereas the peak change in R.sub.2* showed significant association with nodal status, similar association was not found in other perfusion related parameters, such as v.sub.p, BV, BF and MTT. This may be because R.sub.2*-peak.sub.enh is a simple parameter estimate unaffected by any assumptions and sources of errors in kinetic modeling. In particular, correct identification of the AIF is challenging both for DCE- and DSC-MRI in this region, and kinetic parameters derived from AIF deconvolution may therefore be more inaccurate than metrics derived from the raw tissue response. Also, due to a relatively long TE, the first echo used for DCE analysis may be affected by T.sub.2* signal attenuation, especially for high CA concentrations in blood, which may result in an additional errors in the DCE-derived parameters such as K.sup.trans and v.sub.p. However, given the multi-echo acquisition, T.sub.2* signal attenuation in DCE-data may be corrected for by including ΔR.sub.2*-data when estimating the change in R.sub.1.

(57) In conclusion, a dynamic contrast-based multi-echo MRI technique has been implemented and tested in rectal cancer patients. This showed a significant association between the peak change in tumor R.sub.2* during bolus first pass and nodal status, suggesting that DSC-MRI may help to determine N-status in diagnostic rectal cancer staging.

(58) It should be apparent that the foregoing relates only to the preferred embodiments of the present application and the resultant patent. Numerous changes and modification may be made herein by one of ordinary skill in the art without departing from the general spirit and scope of the invention as defined by the following claims and the equivalents thereof.