In-vivo proton magnetic resonance spectroscopy in adnexal lesions.

A resonance peak at 1.3 ppm, caused by the presence of the methylene group in fatty acid chains, was observed in five of the seven malignant ovarian tumors and seven of the eleven benign teratomas ( Figs. 1 , 4 ), as well as in one borderline malignant tumor and the case of salpingitis ( Fig. 2 ). Interestingly, whereas most primary and secondary malignant ovarian tumors showed an intense lipid peak at 1.3 ppm, this same peak was not observed in the MR spectra of any of the six benign ovarian epithelial tumors ( Fig. 3 ). Thus, the incidence of a lipid peak was significantly higher in malignant tumors than in benign epithelial tumors ( p < 0.05). A lipid peak at 5.2 ppm, due to the presence of the olefine group in fatty acid chains, was observed in two cases of the eleven benign teratomas, one of the four endometrioses, and the ectopic pregnancy. The 1 H- MR spectra of benign ovarian epithelial tumors showed no detectable lipid peak at 1.3 or 5.2 ppm ( Tables 1 , 2 ). The lipid/water ratio was determined in 26 cases and was smaller than 0.020 in 21 of these. The lipid/water ratios of benign teratomas showed great variation, and some were higher than those of other adnexal lesions. In addition, all five cases in which the lipid/water ratio was greater than 0.020 were benign teratomas ( p < 0.05).

In our study, 1 H- MR spectroscopic patterns differed markedly according to the histopathologic diagnosis of the adnexal lesions. In fatty acid chains, the presence of the methylene group caused a resonance peak at 1.3 ppm in most patients with malignant ovarian tumors, benign teratomas and salpingitis. However, no lipid peak at 1.3 ppm was found in the 1 H- MR spectra of benign ovarian epithelial tumors or endometrioses. In many previous studies involving in-vivo and ex-vivo MRS of brain tumors and uterine cervical carcinomas, the resonance peak due to fatty acid chains has been suggested as a possible cancer marker ( 12 - 14 ); the results of our study also support this possibility. Interestingly, Gotsis et al. ( 14 ) reported that 1 H- MRS demonstrated the presence of lipids in malignant brain tumors, abscesses and epidermoid cysts, but not in benign brain tumors. They assumed that necrosis and lipid formation might be the result of tumor proliferation without a neovascular network sufficient for its metabolic needs. Kuesel et al. ( 12 ) used ex-vivo 1 H- MRS to evaluate mobile lipid resonance at biopsy or in surgical specimens of cerebral astrocytomas, discovering that in high-grade tumors, 1 H- MRS is able to detect high lipid levels. Rapid cellular turnover in malignant tissue usually leads to considerable cellular death, resulting in cellular necrosis, which appears in in-vivo 1 H- MR spectra as a lipid signal. This was previously observed in a study of human cervical carcinoma using in-vivo 1 H- MRS; invasive cervical carcinoma could be differentiated from normal cervix by a peak at 1.3 ppm, with 89% sensitivity and 57% specificity ( 13 ). In our study, a 1 H- MR spectral peak at 1.3 ppm was detected in malignant or borderline malignant ovarian tumors, with 75% sensitivity (6/8) and 65% specificity (15/23). Because the T2 values of fatty acids and macromolecules are very short, 1 H- MRS using a short TE is required for the detection of these molecules. Owing to overlap in the spectral peaks demonstrated by macromolecules and lipids, the peaks obtained with short TE values (< 40 msec) are not as well resolved as those obtained with TE values greater than 90 msec ( 15 - 17 ). To avoid overlap between a lipid peak and one due to other metabolites at 1.3 ppm, the peak at 5.3 ppm caused by the presence of olefine has sometimes been used to identify the lipid component of a tumor ( 12 ). However, it can be assumed that because the baseline of the spectrum is relatively flat, the peaks occurring at about 1.3 ppm in the short TE MR spectrum of the human ovary are due mainly to the presence in fatty acid chains of the methylene group, and possibly lactate, but not macromolecules. According to a report by Massuger et al. ( 11 ), in-vitro 1 H- NMR studies of ovarian cystic fluids have disclosed that the amount of metabolites such as lactate, methionine, pyruvic acid, and 3-hydroxybutylic acid found in malignant cysts was over six times greater than in benign cysts. In our study, however, the only MR spectral difference between benign and malignant ovarian tumors was the lipid peak at 1.3 ppm. Numerous metabolites, including amino acids, fatty acids, and other small molecules are found in ovarian cysts, but due to their limited concentration (i.e. sensitivity) and intrinsic properties such as T1 and T2 relaxation times, in-vivo 1 H- MRS was able to detect only a limited variety. Thus, in our study, we assumed that the peaks occurring at 1.3 ppm were due mainly to a tumor's lipid component. Evidence which also supported our assumption was the occurrence of a prominent peak at 1.3 ppm in most benign mature cystic teratomas in our study. The combination of long- and short-TE spectra may be useful for evaluating the presence within a tumor of metabolites, lipids, and macromolecules ( 18 ). In studies of brain and uterine cervical tumors, 1 H- MRS using long TE has been used to detect long-T2 lipid components: nonmalignant cells of normal lipid have a short T2 value and are thus apparent only in the spectrum of short TE. Lipid components in a malignant tumor, on the other hand, assumed to have a long T2 (> 400 msec), can be detected by long-TE 1 H- MRS ( 13 , 14 , 19 ). The lipid peak thus detected is therefore attributed to malignant ovarian tissues with long T2 values, and this may indeed imply that the 1 H-MR spectroscopic technique is sensitive to early cellular change. In the case of pelvic MRS examination, signal intensity is influenced by motion artifact caused by respiratory and peristaltic bowel movement; an NMR study reported that the high susceptibility of the volume localization sequence to motion was a major disadvantage of short-TE STEAM 1 H- MRS ( 20 ). Hence we used both short- and long-TE MRS at the beginning of this study, but found that the waveforms of long-TE MR spectra were very irregular and non-informative, probably due to artifact. Long- and short-TE MRS data could not be combined, and only short-TE MR spectra could be analysed. Only a few studies using 1 H- MRS have tried to differentiate between ovarian tumors and other pelvic lesions ( 11 , 13 , 21 ). In order to avoid the broad disturbing spectral resonances caused by protein protons, Massuger et al. ( 11 ) recently analysed in-vitro 1 H- NMR data of deproteinized ovarian cyst fluid samples. The in-vitro 1 H- NMR spectra obtained showed significant differences in metabolite concentrations between benign and malignant ovarian cysts, and significantly higher levels of lactate, isoleucine, valine, 3-hydroxybutylic acid, methionine and alanine were detected in malignant epithelial ovarian cysts. According to them, the elevated metabolite level and branched-chain amino acids found in malignant ovarian tumors most likely result from anaerobic glycolysis and enhanced proteolysis. In our study, however, these peaks were not usually detected by in-vivo 1 H- MR spectroscopy because the minimal concentration detectable by 1.5-T MR instruments is currently tens of millimoles. Although various lesions in our study demonstrated a lipid peak at 1.3 ppm, the lipid/water ratio was useful to differentiate ovarian lesions. As they can be diagnosed easily by the detection of a lipid component at CT or MRI, teratomas demonstrated characteristically greater lipid/water ratios than other ovarian lesions; all our five cases in which the lipid/water ratio was greater than 0.020 were teratomas. This study has several limitations, one of which is that 1 H- MR spectroscopic and radiologic findings were not compared. Another is the small number of cases studied. In order to establish indications of in-vivo 1 H- MR spectroscopic diagnosis, further investigation involving a large number of cases is required. In conclusion, various ovarian lesions can be distinguished by careful disclosure of in-vivo 1 H- MR spectra, and in-vivo 1 H- MRS is potentially able to discriminate between malignant and benign tumors. The presence of a lipid peak at 1.3 ppm is a useful indicator of malignancy.

Thirty-one patients aged 22-69 (mean ± SD, 40 ± 14) years with surgically proven adnexal lesions were included in this study. There were seven primary or secondary malignant ovarian tumors (2 papillary serous cystadenocarcinomas, 1 mucinous cystadenocarcinoma, 1 sarcomatoid carcinoma, and 3 metastatic adenocarcinomas), one borderline malignant mucinous cystic tumor, six benign epithelial tumors (3 mucinous cystadenomas, 1 serous cystadenoma, 1 serous cystadenofibroma, and 1 benign mixed epithelial tumor), eleven benign teratomas, four endometrioses, one ectopic pregnancy, and one case of salpingitis ( Table 1 ). The maximal diameter of the lesions ranged from 4 to 23 cm, which in all cases was greater than the localization voxel for 1 H- MRS. Prior to 1 H- MRS, pelvic CT and/or MRI was performed in all patients, none of whom has undergone treatment or biopsy. For in-vivo 1 H- MRS, a 1.5-T whole-body GE MRI/MRS system (Signa Horizon; General Electric Medical Systems, Milwaukee, Wis., U.S.A.) equipped with active shielded gradients was used. Prior to 1 H- MRS, pelvic CT or MRI findings were reviewed by a radiologist to determine the location of adnexal lesions. MR imaging using a fast multiplanar spoiled gradient-echo (FMPSPGR) sequence (TR/TE= 100/1.5, NEX=1, matrix size=256×128, slice thickness=8.0 mm, interslice gap=2.0 mm) was performed in which the localization voxel was placed on the adnexal lesion. Voxel size was approximately 8 cm 3 (2×2×2cm), and the voxel was located safely within the lesion in order to reduce contamination from adjacent normal tissue. For all spectra, a STEAM (stimulated echo acquisition method) localization sequence combined with a three-pulse CHESS (chemical shift selective sequence) to suppress the H 2 O signal was used, and the acquisition parameters were as follows: TR=3000 ms, TE=30 ms, number of scans=128, and number of excitations=1. During spectroscopic signal acquisition, the subjects respired quietly without interruption. Respiratory gating was not used in this study. Post-processing involving the use of a SUN SPARC 20 workstation with Spectral Analysis/General Electric (SA/GE) software; incorporated in this were low-frequency filtering of residual water signal, apodization by 0.5 Hz of exponential line broadening, zerofilling of 8000, Fourier transformation, and lorenzian-to-gaussian transformation. 1 H- MR spectra both with and without water-peak suppression were obtained, and all spectra were analysed with particular attention to the presence of lipid peaks at 1.3 and 5.2 ppm, and a water peak at 4.77 ppm. Lipid/water ratios were calculated by integrating the peak area of the lipid peak at 1.3 ppm and water peaks in water-unsuppressed spectra. In spectra that demonstrated a dominant lipid peak at about 1.3 ppm, the highest peak point ± 0.3 ppm of both the lipid and water peak was integrated. In 1 H- MR spectra that showed no dominant lipid peak, a range of 1.0-1.6 ppm was integrated in order to determine lipid content. The chi-square test was used to analyse spectroscopic data, and difference rankings between means at a statistically significant level were thus established. A p value of 0.05 was considered statistically significant.