Hepatobiliary Imaging
MR Imaging of Hepatic Metastases Caused by Neuroendocrine Tumors: Comparing Four Techniques
Audio Available | 
Share
OBJECTIVE. The aim of our prospective study was to assess the MR imaging characteristics of hepatic metastases of neuroendocrine tumors and to determine the optimal MR sequence for their detection.
SUBJECTS AND METHODS. Thirty-seven consecutive patients with liver metastases from neuroendocrine tumors underwent 1.5-T MR imaging of the liver comprising T2-weighted fast spin-echo with respiratory monitoring, breath-hold T2-weighted single-shot fast spin-echo, and T1-weighted gradient-recalled echo sequences before and after the injection of gadoterate dimeglumine. Images were reviewed independently by three observers for the number, location, and pattern of signal and enhancement of metastases.
RESULTS. A total of 359 metastases were detected, 279 on T2-weighed fast spin-echo, 231 on T2-weighed single-shot fast spin-echo, 272 on unenhanced T1-weighted, 322 on hepatic arterial phase, and 228 on portal venous phase images. Hepatic arterial phase images revealed the greatest number of metastases in 70% of patients, including 35 metastases seen only on this sequence, and was significantly superior to the unenhanced T1-weighted and portal venous phase sequences (p < 0.01). The lesion-to-liver contrast was significantly greatest with T2-weighed fast spin-echo sequences. The enhancement patterns of metastases were predominantly hypervascular, hypovascular, peripheral with progressive fill-in, and delayed in, respectively, 27, four, four, and two patients. Most metastases with peripheral enhancement and progressive fill-in were heterogeneous on T2-weighted images and were without globular peripheral enhancement.
CONCLUSION. Hepatic metastases of neuroendocrine tumors had a typical hypervascular pattern in 73% of patients. Hepatic arterial phase and fast spin-echo T2-weighed sequences are the most sensitive.
| Introduction |   |
|---|
Neuroendocrine tumors are rare malignancies that arise from the Kulchitsky's enterochromaffin cells and that are characterized by positivity for chromogranin A and synaptophysin antibodies at immunohistochemical analysis. Primary tumors can arise in various organs but occur predominantly in the gastrointestinal tract, pancreas, and lung. Most often, gastrointestinal neuroendocrine tumors are small tumors, and the primary tumor is revealed through the detection of metastases. The most common metastatic sites are the liver, lymph nodes, bone, lung, and peritoneal cavity. Hepatic metastases and the degree of liver involvement are considered major prognostic factors for survival in patients with neuroendocrine tumors [1,2,3]. Even if neuroendocrine tumors are rare, the accurate staging of hepatic disease is of paramount importance because long-term survival is possible [4].
Hepatic metastases from neuroendocrine tumors are often highly vascular. Previous reports have documented the value of helical or multidetector CT in the radiologic assessment of hypervascular hepatic tumors because those techniques allow multiphase imaging during a single bolus administration of contrast material [5,6,7,8]. To our knowledge, only a few studies have studied neuroendocrine tumor metastases on MR imaging [9, 10], but those studies did not evaluate prospectively the value of various MR pulse sequences in terms of lesion detection. The objective of our single-center prospective study was to assess the MR imaging characteristics of neuroendocrine tumor—derived hepatic metastases and to determine the optimal MR pulse sequences, the value of gadolinium enhancement, and the best phase in the dynamic study for detecting these lesions.
| Subjects and Methods | |
|---|
From September 2000 to September 2001, 48 consecutive patients having a neuroendocrine tumor who were referred to our center underwent MR imaging of the liver for screening of potential hepatic disease and were enrolled in this prospective study. Institutional review board approval and informed consent of the patient were not obtained because all MR imaging was performed during the usual follow-up of the patients. Of these patients, 11 had no evidence of focal hepatic metastases on any of the MR images and were excluded from further analysis. The remaining 37 patients—20 women and 17 men with a mean age of 59 years (range, 29-76 years)—composed the study population. Primary sites of neuroendocrine tumors were the small bowel (n = 12), the pancreas (n = 9), the lung (n = 6), the colon and rectum (n = 3), the appendix (n = 2), the thymus (n = 1), and unknown (n = 4). Seventeen patients had received treatment, including segmental hepatectomy (n = 4), systemic chemotherapy (n = 7), and somatostatin analogs (n = 10), before MR imaging. Hepatic metastases were proven at histologic analysis after hepatectomy (n = 7), open surgical biopsy (n = 11), percutaneous core biopsy (n = 7), and somatostatin receptor-positive focal liver lesions at scintigraphy (n = 8), and after the appearance of new hepatic lesions in a patient who had previously undergone surgery for liver metastases from a carcinoid tumor (n = 4). A biopsy specimen was obtained from only one lesion per patient, and patients with multiple lesions exhibiting a similar appearance were presumed to have multifocal metastases of the same histologic type as that of the biopsy specimen.
MR imaging was performed with a 1.5-T whole-body imager (Signa LX; General Electric Medical Systems, Milwaukee, WI). All MR images were acquired in the axial plane using a phased array body multicoil. Slice thickness was 7 mm, with a 2-mm intersection gap for all pulse sequences. Fat-suppressed T2-weighted images were obtained with a respiratory-triggered fast spin-echo sequence using the following parameters: effective TR range/effective TE, 4000-8000/102; echo-train length, 16; signals acquired, 4; interecho spacing, 10 msec; matrix, 256 × 256; received bandwidth, 31.25 kHz; field of view, 280-400 mm; respiratory trigger point, 20%; trigger window, 40%; and gradient moment nulling in the frequency-encoding direction. Saturation bands superior and inferior to the imaging volume were used to attenuate flow-related artifacts throughout MR imaging. Breath-hold single-shot fast spin-echo images were obtained using the following parameters: effective TR/effective TE, infinite/90; half-Fourier acquisition; echo-train length, 104; signal acquired, 1; matrix, 256 × 256; received bandwidth, 62.5 kHz; and field of view, 40 cm. Dynamic contrast-enhanced MR images were obtained with an initial delay of 10 sec and then at three consecutive 30-sec intervals and at 5 min after the initiation of a bolus injection of 0.1 mmol/kg of gadoterate dimeglumine (Dotarem; Guerbet, Aulnay-sous-Bois, France) into the antecubital vein. T1-weighted sequences were acquired with fast multiplanar spoiled gradient-recalled echo imaging using the following parameters: TR/TE, 125/4.2; flip angle, 60°; signal acquired, 1; matrix, 512 × 256; received bandwidth, 62.5 kHz; field of view, 40 cm; and a 25-sec acquisition time in a single breath-hold.
Signal intensity values for hepatic lesions, healthy liver, and background were obtained with regions of interest that were manually delimited by a radiologist in the same locations for all MR sequences for a given patient. The signal intensity in the liver was measured in areas devoid of large vessels and prominent artifacts. To minimize the difference in signal intensity caused by the near-field effect of the torso coil, the regions of interest for the liver and the hepatic metastases were located at the same vertical distances from the ventral-side surface coils on the same scan. Measurements were made of one representative metastasis for each patient. A circular region of interest was drawn for each metastasis encompassing as much of the lesion as possible. Noise was defined as the standard deviation of background signal intensity and was measured outside the anterior abdominal wall in the phase-encoding direction. The lesion-to-liver contrast-to-noise ratio was calculated for each sequence in the following manner: lesion-to-liver contrast-to-noise ratio = (SIlesion-SIliver)/SDB, where SIlesion and SIliver are the signal intensities of the hepatic lesions and the liver, respectively, and SDB is the standard deviation of the signal intensity of the background. Only the absolute magnitude of the lesion-to-liver contrast-to-noise ratio was taken into account because it correlates with lesion visibility.
The MR images of each patient were separated into five sets corresponding to five different sequences: single-shot fast spin-echo T2-weighted sequence, fast spin-echo T2-weighted sequence, unenhanced T1-weighted sequence, hepatic arterial phase sequence, and portal venous phase sequence. Hepatic arterial phase and portal venous phase images corresponded to the enhanced T1-weighted images acquired with a delay of, respectively, 10 and 70 sec after the initiation of contrast medium injection. The five sets of images were reviewed independently by three observers (two experienced senior radiologists and one junior radiologist who was not experienced in MR imaging of the liver). The order of the five sets of images was randomized before review. Images were all interpreted on a digital workstation (Pathspeed; General Electric Medical Systems) with total freedom for window and level adjustments and for the magnification of each image at the time of analysis. The three radiologists were unaware of clinical and biologic findings or findings of any imaging studies concerning the patients. However, the radiologists knew that all patients had a neuroendocrine tumor and were being examined to evaluate the extent of hepatic metastases.
For each set of images, the three reviewers described the number, size, location (Couinaud segment), signal intensity, and pattern of enhancement of each metastasis. Only presumed metastatic lesions for which the confidence rate was high were taken into account. When more than 15 metastases were depicted, 16 were considered for the statistical analysis. The signal intensity pattern of metastases was compared with that of the surrounding liver parenchyma and classified as hyperintense, hypointense, or isointense on unenhanced T1-weighted, T2-weighted, hepatic arterial phase, and portal venous phase images. We distinguished four enhancement patterns: predominantly hypovascular, hypervascular with early (seen at the hepatic arterial phase) and intense global enhancement, delayed enhancement (best seen on 5-min images), and peripheral enhancement with progressive centripetal fill-in. The number of metastases detected on all sequences in each patient was then determined independently by the three reviewers. The number of metastases seen on only one pulse sequence was recorded for each patient.
The correlation coefficient was calculated to assess the degree of agreement between each combination of two reviewers. Wilcoxon's signed rank test for paired data was used to compare pulse sequences. A p value of less than 0.05 was considered significant.
| Results | |
|---|
The lesion-to-liver contrast-to-noise ratios obtained with the five sequences are reported in Table 1. The T2-weighted fast spin-echo pulse sequence with fat suppression showed the greatest lesion-to-liver contrast-to-noise ratio, which was significantly greater (p < 0.02) than those obtained with unenhanced T1-weighted images, hepatic arterial phase, and portal venous phase images. No significant difference was found in lesion-to-liver contrast-to-noise ratios between fast spin-echo and single-shot fast spin-echo T2-weighted MR images (p = 0.313). No significant difference was found in lesion-to-liver contrast-to-noise ratios between hepatic arterial phase and portal venous phase images (p = 0.124) or between hepatic arterial phase and unenhanced T1-weighted images (p = 0.153).
The values of the correlation coefficient between each pair of reviewers are summarized in Table 2. Good interobserver agreement was obtained for the number of metastases seen on each sequence. The best agreement was achieved when all sequences were combined, whereupon concordance was lowest for the portal venous phase. The total number of metastases detected by each reviewer for each sequence is shown in Table 3. A total of 359 metastases, ranging from 4 to 85 mm (median, 14 mm) in maximum diameter, were depicted. A trend was seen toward increased sensitivity with the hepatic arterial phase compared with other sequences. The hepatic arterial phase sequences depicted more hepatic metastases (322/359, 90%) than did the fast spin-echo T2-weighted sequence with fat suppression (279/359, 78%), although no significant differences were observed between these two sequences (p = 0.29). The hepatic arterial phase was significantly better than unenhanced T1-weighted (272/359, 76%; p < 0.01), single-shot fast spin-echo T2-weighted (231/359, 64%; p < 0.001), and portal venous phase (228/359, 64%; p < 0.001) sequences (Fig. 1A,1B,1C). Fast spin-echo T2-weighted images with fat suppression depicted significantly more hepatic metastases than did single-shot fast spin-echo T2-weighted images (p < 0.01). No significant difference was found in sensitivity between unenhanced T1-weighted images and portal venous phase images except by one reviewer (p < 0.05). Thirty-five metastases (10%) were seen only on hepatic arterial phase images; 11 metastases (3%), only on fast spin-echo T2-weighted images; eight metastases (2%), only on unenhanced T1-weighted images; three metastases (1%), only on single-shot fast spin-echo T2-weighted images; and four metastases (1%), only on portal venous phase images.
 View larger version (151K) | Fig. 1A. —Transverse MR images of multifocal hepatic metastases from neuroendocrine tumor best depicted on hepatic arterial phase in 63-year-old woman. Fat-suppressed T2-weighted respiratory-triggered fast spin-echo image (TR/TE, 6315/100) shows few metastases (arrows). |
 View larger version (133K) | Fig. 1B. —Transverse MR images of multifocal hepatic metastases from neuroendocrine tumor best depicted on hepatic arterial phase in 63-year-old woman. Arterial phase contrast-enhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo image (150/4.2) shows metastases depicted on A (arrows) and numerous additional metastases (arrowheads) as multifocal nodular enhancement corresponding to hypervascular metastases. |
 View larger version (128K) | Fig. 1C. —Transverse MR images of multifocal hepatic metastases from neuroendocrine tumor best depicted on hepatic arterial phase in 63-year-old woman. Venous phase contrast-enhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo image (150/4.2) does not show any metastases. |
On a patient-per-patient basis, the maximum number of metastases was found on hepatic arterial phase images in 70% of patients, on fast spin-echo T2-weighted images in 55% of patients, on unenhanced T1-weighted images in 54% of patients, on single-shot fast spin-echo T2-weighted images in 38% of patients, and on portal venous phase images in 36% of patients (Table 4). The maximum number of metastases was depicted by only one sequence in 14 patients (38%), by two sequences in five patients (14%), by three sequences in six patients (16%), by four sequences in three patients (8%), and by all five sequences in nine patients (24%). When two sequences were combined, the maximum number of metastases was detected in 30 patients (81%) when hepatic arterial phase and fast spin-echo T2-weighted sequences were combined, in 29 patients (78%) when hepatic arterial phase and unenhanced T1-weighted sequences were combined, and in 26 patients (70%) when unenhanced T1-weighted and fast spin-echo T2-weighted sequences were combined.
Table 5 shows the classification of metastases on each of the sequences according to their relative signal intensity. Most metastases were hypointense on unenhanced T1-weighted images (93-94%), moderately hyperintense (85-86%) or strongly hyperintense (close to the signal of liquid) on T2-weighted images, and hyperintense on hepatic arterial phase and portal venous phase images (80% and 70%, respectively). The enhancement pattern was hypervascular in 27 patients (289 metastases) (Figs. 1A,1B,1C and 2A,2B,2C,2D), hypovascular in four patients (20 metastases) (Fig. 3A,3B,3C), peripheral with progressive fill-in in four patients (34 metastases) (Fig. 4A,4B,4C,4D), and delayed in two patients (six metastases). The enhancement pattern depicted in 34 metastases with peripheral and progressive fill-in was a peripheral rim of enhancement without a globular pattern. The metastases corresponded to heterogeneous lesions on T2-weighted images exhibiting areas of high (waterlike) and moderate signal intensity (Fig. 4A,4B,4C,4D).
 View larger version (178K) | Fig. 2A. —Characteristic pattern of hypervascular hepatic metastases from neuroendocrine tumor in 54-year-old woman. Transverse fat-suppressed T2-weighted respiratory-triggered fast spin-echo MR image (TR/TE, 7500/100) shows multiple hyperintense metastases with high lesion-to-liver contrast. |
 View larger version (169K) | Fig. 2B. —Characteristic pattern of hypervascular hepatic metastases from neuroendocrine tumor in 54-year-old woman. Transverse unenhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo MR image (150/4.2) shows multiple hypointense metastases. |
 View larger version (175K) | Fig. 2C. —Characteristic pattern of hypervascular hepatic metastases from neuroendocrine tumor in 54-year-old woman. Transverse arterial phase contrast-enhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo MR image (150/4.2) shows early and marked enhancement of metastases. |
 View larger version (167K) | Fig. 2D. —Characteristic pattern of hypervascular hepatic metastases from neuroendocrine tumor in 54-year-old woman. Transverse venous phase contrast-enhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo MR image (150/4.2) shows decreased enhancement of metastases (washout) and decrease in lesion-to-liver contrast and in number of metastases. |
 View larger version (191K) | Fig. 3A. —Hypovascular metastases from neuroendocrine tumor in 58-year-old man. Transverse breath-hold T2-weighted single-shot fast spin-echo MR image (TR/TE, 48569/92.3) shows no obvious hepatic metastases. |
 View larger version (195K) | Fig. 3B. —Hypovascular metastases from neuroendocrine tumor in 58-year-old man. Transverse arterial phase contrast-enhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo MR image (150/4.2) shows numerous hypointense metastases (arrows) with no early enhancement. |
 View larger version (187K) | Fig. 3C. —Hypovascular metastases from neuroendocrine tumor in 58-year-old man. Transverse venous phase contrast-enhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo MR image (150/4.2) shows increase in lesion-to-liver contrast resulting from no enhancement of hepatic metastases. |
 View larger version (181K) | Fig. 4A. —Large hepatic metastasis from neuroendocrine tumor mimicking atypical hemangioma in 68-year-old man. Transverse fat-suppressed T2-weighted respiratory-triggered fast spin-echo MR image (TR/TE, 5454/100) shows extremely hyperintense lesion (arrow) with signal intensity close to that of liquid and similar to expected signal intensity of hemangioma. However, lesion is heterogeneous, with central foci exhibiting moderately high signal intensity. |
 View larger version (187K) | Fig. 4B. —Large hepatic metastasis from neuroendocrine tumor mimicking atypical hemangioma in 68-year-old man. On transverse breath-hold T2-weighted single-shot fast spin-echo MR image (48569/92.3), lesion (arrow) had lower signal intensity than on fast spin-echo T2-weighted image, thus making it different from signal intensity of water. |
 View larger version (153K) | Fig. 4C. —Large hepatic metastasis from neuroendocrine tumor mimicking atypical hemangioma in 68-year-old man. Transverse arterial phase contrast-enhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo MR image (160/4.2) shows peripheral rim of enhancement (arrow) similar to that of hemangioma but without globular pattern characteristic of benign hemangioma. |
 View larger version (140K) | Fig. 4D. —Large hepatic metastasis from neuroendocrine tumor mimicking atypical hemangioma in 68-year-old man. Transverse venous phase contrast-enhanced breath-hold T1-weighted fast multiplanar spoiled gradient-recalled echo MR image (160/4.2) shows progressive centripetal fill-in enhancement (arrow) similar to that found in hemangioma. |
| Discussion | |
|---|
Neuroendocrine liver metastases are usually highly vascularized, with their blood supply arising from the hepatic artery. In theory, hepatic arterial phase images should therefore provide the best results in the detection of these metastases regardless of the imaging technique used (CT or MR imaging). In a previous CT study of hepatic metastases from neuroendocrine tumors, Paulson et al. [6] found 30% of additional metastases during the hepatic arterial phase; and in 6% of patients, hepatic metastases were seen exclusively on the hepatic arterial phase. Breath-hold dynamic contrast-enhanced T1-weighted MR imaging allows multiphase imaging of the entire liver with sequences repeated every 30 sec and has become a standard method for the detection and characterization of hypervascular hepatic tumors, be they benign (focal nodular hyperplasia, adenoma) or malignant (hepatocellular carcinoma) [11, 12]. To the best of our knowledge, few studies reporting on a limited number of patients have focused on MR imaging of hepatic metastases from neuroendocrine tumor. De Santis et al. [10] reported on six patients without comparing MR sequences when detecting metastasis. Bader et al. [9] reported on 16 patients studied retrospectively over 9.5 years and showed that 94% of metastases were hypervascular and 15% were seen only on immediate gadolinium-enhanced images. Finally, two studies have shown that, unlike the enhancement shown for other liver metastases, mangafodipir can enhance hepatic metastases from neuroendocrine tumors, probably because of the increased arterial flow and high metabolic activity in those lesions [13, 14].
In our study, the hepatic arterial phase was the sequence that revealed the greatest number of metastases; it was significantly superior to unenhanced and portal venous phase images. The hepatic arterial phase revealed 90% of metastases, including the 35 metastases visualized exclusively on this sequence. This was also the sequence on which the maximum number of metastases were detected in 70% of patients. However, even if the hepatic arterial phase images had not been obtained, no patient with metastases would have been missed; and furthermore, no significant difference was seen between the hepatic arterial phase images and the fat-suppressed fast spin-echo T2-weighted images in the detection of metastases. With the greatest lesion-to-liver contrast-to-noise ratio (significantly greater than hepatic arterial phase), the fat-suppressed fast spin-echo T2-weighted sequence was the most useful to combine with the hepatic arterial phase because that combination allowed the detection of the maximum number of metastases from neuroendocrine tumors in more than 80% of patients. The unenhanced T1-weighted sequence showed good efficiency by depicting 76% of metastases, and it ranked second when combined with the hepatic arterial phase sequences for depicting the most metastases in 78% of patients. However, the unenhanced T1-weighted pulse sequence exhibited the lowest lesion-to-liver contrast-to-noise ratio, which probably accounts for the widest interobserver discrepancies in the number of metastases depicted (Table 3), because it is difficult to visualize metastases with a high confidence on those sequences. The portal venous phase images were the least informative, depicting 63% of metastases; this sequence was superior to others in only four patients who had metastases with hypovascular enhancement.
When T2-weighted respiratory-triggered fast spin-echo MR imaging became the standard approach for imaging of the liver, single-shot sequences were developed to decrease imaging time and motion artifacts. A single-shot sequence provides high contrast for tissues with a very long T2 relaxation time (cyst, hemangioma) and therefore is often used with a long TE (>800 msec) for MR cholangiopancreatography. Previous reports have shown a similar performance but a reduced lesion-to-liver contrast-to-noise ratio for solid lesions with half-Fourier single-shot acquisition compared with conventional fast spin-echo sequences [15,16,17]. In our study, as regards lesion detection, the performance and the lesion-to-liver contrast-to-noise ratio of fast spin-echo images were significantly superior to those of single-shot fast spin-echo T2-weighted images despite the hypervascularized nature of these metastases. Two main reasons have been reported to explain the poor diagnostic capability of single-shot fast spin-echo imaging for solid lesions: the magnetization transfer contrast effect is considered responsible for a decrease in signal intensity of solid lesions [18], and the T2 filtering effect has been reported to lead to image blurring [17].
We found that metastases had a typical hypervascular enhancement in 73% of patients, an atypical hypovascular or delayed enhancement in 16% of patients, and a peripheral enhancement with progressive fill-in mimicking hemangioma in 11% of patients. Histologic proof was obtained in all four patients having 34 metastases with peripheral enhancement and progressive fill-in. An atypical MR imaging pattern of hemangioma with heterogeneity on T2-weighted images and without a globular pattern of peripheral enhancement was found in all these pseudoangiomatous metastases as described by others [19]. The pseudoangiomatous metastases had lower signal intensity on single-shot fast spin-echo T2-weighted pulse sequence than on the fast spin-echo T2-weighted pulse sequence because the lesion-to-liver contrast-to-noise ratio for solid lesions was lower with single-shot fast spin-echo than with the fast spin-echo pulse sequence. Thus, single-shot fast spin-echo is a useful sequence to make the distinction between metastases with peripheral and progressive fill-in and hemangiomas, especially in the case of small lesions in which the peripheral enhanced rim characteristic of metastases is difficult to distinguish from the globular peripheral enhancement characteristic of hemangiomas.
Our study is limited in certain respects. Although metastatic disease to the liver was proven in all patients, a detailed lesion-by-lesion histopathologic analysis was not possible because most of the patients in this series had metastases from the neuroendocrine tumor and therefore did not undergo hepatic resection. Seven of our patients had previously received chemotherapy. In these seven patients, the enhancement pattern of metastases was hypervascular in six and delayed in one. However, chemotherapy may alter the vascularity of metastases, thereby lowering the performance of enhanced hepatic arterial phase imaging [20]. In our study, the contrast injection rate was low (3 mL/sec), but it was chosen so that all patients would have a similar rate, even those with poor venous access. However, a low contrast injection rate can also limit the sensitivity of hepatic arterial phase images. The maximum number of metastases taken into account was limited to 15, which probably underestimates the contribution of hepatic arterial phase images, especially in cases of minute hepatic metastases, most of which were seen only on the hepatic arterial phase images.
In conclusion, hepatic metastases from neuroendocrine tumors had a typical hypervascular pattern in 73% of patients. The combination of hepatic arterial phase and fat-suppressed fast spin-echo T2-weighted images depicted 80% of hepatic metastases derived from neuroendocrine tumors, and those sequences could be considered the basic MR imaging sequences in these patients.
Address correspondence to C. Dromain.
We thank Lorna Saint Ange for editing.
| References | |
|---|
