Background

The most common primary brain malignancies are gliomas (approximately 80%) [1]. The incidence and 5‑year survival rate are 1–4 per 100,000 and 4–40%, respectively. The grade of the central nervous system tumors, classified by the World Health Organization (WHO), determines the survival rate and management plan. According to this classification, gliomas are divided into four grades. Grades 1 and 2 are considered low-grade glioma (LGG) and grades 3 and 4 are considered high-grade glioma (HGG; [2, 3]). If LGGs are resected at the time of diagnosis, the 5‑year survival rate is 94% and 51.6% for grades 1 and 2, respectively [4,5,6]. The treatment and prognosis for high-grade gliomas vary greatly. They usually require radiation therapy and chemotherapy after surgery, and the 5‑year survival rate is only 25.2% and 8.9% for grades 3 and 4, respectively. [5,6,7].

Although histological tests are the gold standard for the diagnosis of glioma, they have certain limitations such as high invasiveness, bleeding and infection, and likelihood of sampling errors, and they are prone to inter- and intra-neuropathologist variability. Complications can arise depending on the location and grade of the tumors [8]. Imaging has the advantage that histopathological tumor changes can be visualized indirectly and non-invasively. Magnetic resonance imaging (MRI) is the imaging method of choice for the detailed characterization of brain tumors, since it provides valuable information on tumor composition, anatomical location, edema, hemorrhage, adjacent brain tissue, and mass effect [9]. Additionally, contrast enhancement indicates neovascularization or disruption of the blood–brain barrier, which is an important feature of HGG. Despite these advantages, conventional MRI has a limited role in grading gliomas due to its high rate of false-positive results (tumor heterogeneity; [9, 10]).

Previous studies have shown that increased levels of free (mobile) lipids are associated with higher grades in brain glioma. Magnetic resonance spectroscopy (MRS) is a noninvasive test approved as a noninvasive modality for assessing lipid concentration and tumor grade [10, 11]. Chemical shift imaging is a gradient-echo (GRE) MRI technique in which in-phase and opposed-phase (IOP) sequences are used clinically to spot lipids in renal angiomyolipoma [12], adrenal adenoma [13], bone marrow involvement [14], and whole-body fat quantification [15]. It could also be useful for detecting lipids in brain gliomas [16]. Elevated levels of HGG lipids may indicate greater loss of signal in the opposed-phase sequences [8, 17, 18]. The advantages of chemical shift IOP sequences are the shorter acquisition time, non-sensitivity to adjacent bone artifacts, wide applicability in MRI devices, general evaluation of the abundance of lipids, and no need for high-priced software applications and expertise [11, 19].

In the current study, we evaluated the in-phase and opposed-phase sequences for distinguishing HGG from LGG and their correlation with MRS results.

Methods

Participants

This retrospective observational study of prospectively collected preoperative imaging data was approved by the research registrar of Mashhad University of Medical Sciences and was conducted between September 2020 and October 2022 at Ghaem Hospital in Mashhad, Iran. Written consent was obtained from all participants. This study was conducted with patients with brain tumors who were referred to our Brain MRS Center. Inclusion criteria were patient age ≥ 18 years, untreated brain tumor, and MRS images of acceptable quality. The gold standard for diagnosis was based on the World Health Organization (WHO) classification of gliomas [2]. Brain tumors were divided into low-grade (WHO grades 1 and 2) and high-grade (WHO grades 3 and 4) groups [2, 3]. All histopathological examinations were derived from the total or partial resection. Once pathology results were generated, patients with diagnoses other than glioma were excluded from the study. Figure 1 illustrates the selection process.

Fig. 1
figure 1

Flowchart of the selection process. GE gradient-echo sequence, IOP in-phase and opposed-phase sequence

Full size image

Imaging and data acquisition

A standard tumor protocol included axial FLAIR (TR, 9000 ms; TE, 114 ms; section thickness, 5 mm), T2-weighted (TR, 4000 ms; TE, 90 ms; section thickness, 5 mm), and T1-weighted (TR, 500 ms; TE, 14 ms; section thickness, 5 mm) images; coronal T1- and T2-weighted images; postcontrast T1-weighted, diffusion weighted imaging (DWI) with derived apparent diffusion coefficient (ADC; TR, 3200 ms; TE, 68 ms; section thickness, 5 mm); and single-voxel spectroscopy (SVS) images (TR, 1500 msec; TE, 35 msec, and a VOI of 1 × 1 × 1 cm3), which were acquired using a 1.5‑T MRI scanner (Ingenia®, Philips [Amsterdam, the Netherlands]) that was provided with a head coil.

Before administration of the contrast agent, extra chemical shift GRE IOP sequences were acquired. Chemical shift IOP parameters were as follows: repetition time of 160 ms; echo time of 2.4 ms for the in-phase and 4.2 ms for the opposed-phase; flip angle, 80°; section thickness, 5 mm; number of averages = 1; field of view, 250 × 250 mm2; matrix size, 256 × 256. On SVS, free lipid concentration was estimated based on the area under the curve (AUC) at 0.9 and 1.3 ppm peaks. Contrast-enhanced series were obtained after intravenous administration of gadoterate meglumine (Dotarem®, Guerbet [Aulnay sous Bois, France]) at a dose of 0.2 mL/kg and a flow rate of 2 mL/s.

An expert radiologist examined the in-phase and opposed-phase images. Three 20–30-mm2 oval regions of interest (ROIs) were placed manually in the solid component in the in-phase and opposed-phase images, and the average signal intensity (SI) values were determined. The ROI was taken from the same solid area from which the MRS sequence was obtained. The solid part of the tumor is defined as the part of the tumor that is isointense on T1-weighted (T1W) images and presents an iso- to hyperintense signal on T2-weighted (T2W) images. The same “solid component” was used on both SVS and IOP imaging analyses. Based on the presence of the decreased SI area in the opposed-phase sequences by visual evaluation, we found the most suitable location for the ROI. The same ROIs were then used consistently across both sequences (Fig. 2).

Fig. 2
figure 2

A 60-year-old man with high-grade glioma. a Magnetic resonance spectroscopy (MRS) showed elevated choline and low N-acetylaspartate (NAA) levels with lactate peak. b Three regions of interest were placed on each chemical shift sequence (right: opposed-phase, left: in-phase) to calculate the mean signal intensities and corresponding signal loss ratio

Full size image

The signal loss ratio (SLR) was calculated using the following formula for estimating intravoxel fat content:

$$\mathrm{SLR}_{\text{tumor}}=(\mathrm{SI}_{\text{In phase}}-\mathrm{SI}_{\text{Opposed phase}})/\mathrm{SI}_{\text{In phase}}$$

Both quantitative (concentration: area under the curve) and qualitative (peak presence: yes/no) MRS measurements of lipid-lactate and free lipid were extracted. Demographic information such as age and gender were extracted from the files.

Spectroscopic data quality was systematically evaluated across all examinations. Consistency and reliability were ensured by assessing factors such as noise levels, calibration, and precision.

Data analysis

Based on previous work, a sample size of 20 was estimated (α error: 0.05 and power: 80%) [17]. Numerical variables are expressed as mean ± standard deviation (SD). Qualitative variables are expressed as numbers and percentages. The independent sample t test and chi-square test were used to compare parameters between the two groups. The correlation between the average SLR and the concentrations of different lipids was assessed with the Pearson correlation test. Receiver operating characteristic (ROC) curve analysis was performed to find the optimal threshold for the SLR and to assess its diagnostic performance. The diagnostic power of the test was graded as follows: 0.5–0.6, fail; 0.6–0.7, poor; 0.7–0.8, fair; 0.8–0.9, good, 0.9–1, excellent [20]. We used IBM SPSS Statistics for Windows, Version 25.0 (IBM Corp., Armonk, NY, USA) for the data analysis. Statistical significance was set at p < 0.05.

Results

Participants

This study was performed with 33 patients: 20 patients were included in the low-grade and 13 in the high-grade group. The patient ages ranged from 18 to 75 years (43.36 ± 15.192). The study comprised 18 (54.5%) male patients and 15 (45.5%) female patients. The correlation between SLR, age, and gender was not statistically significant (p = 0.051 and 0.382, respectively).

Tumor characteristics

In order of frequency, the most common tumor location was frontal lobe (39.4%), temporal lobe (33.3%), parietal lobe (12.1%), pons (9.1%), and thalamus (6.1%).

The mean lipid-lactate concentration in HGG and LGG was 61.41 ± 89.7 and 3.51 ± 7.49, respectively (p = 0.007). The mean free lipid concentration in HGG and LGG was 22.13 ± 18.6 and2.07 ± 2.76, respectively (p < 0.001). Both showed statistically significant differences.

A lipid-lactate peak was present in 14 tumors. The mean SLR of present and non-present peaks was 3.73 ± 2.2 and 1.47 ± 1.63, respectively. A comparison of the SLR between the two groups of present and non-present peaks showed a statistically significant difference (p = 0.004). A mobile lipid peak was present in 12 tumors. The mean SLR of the present and non-present peaks was 3.64 ± 2.3 and 1.74 ± 1.82, respectively. A comparison of the SLR between the two groups of present and non-present free lipid showed a statistically significant difference (p = 0.024).

The mean SLR in the HGG and LGG groups was 3.66 ± 2.12 and 1.63 ± 1.86, respectively. There was a significant difference in the SLR value of the two groups (p = 0.01). According to the results of a Pearson test, there was a statistically significant correlation between lipid-lactate (0.48, p = 0.004) and free lipid (0.44, p = 0.009) concentrations on MRS with SLR. Details of the patient and tumor characteristics and the comparison of the two groups are shown in Table 1.

Table 1 Patient and tumor characteristics
Full size table

Diagnostic performance

The ROC analysis of lipid-lactate concentration showed an AUC of 0.96 and an optimal cut-off of 7.74, leading to a sensitivity of 100% and a specificity of 90%. The AUC for lipid concentration was 0.97 and the optimal cut-off was 5.2, leading to a sensitivity of 100% and a specificity of 90%. The AUC for SLR was 0.75 and the optimal cut-off was 2.88, leading to a sensitivity of 69% and a specificity of 85% (Table 2 and Fig. 3).

Table 2 Diagnostic performance
Full size table
Fig. 3
figure 3

Receiver operating characteristic (ROC) analysis of lipid lactate concentration, lipid concentration, and signal loss ratio (SLR) for distinguishing between low-grade and high-grade gliomas. AUC area under the curve

Full size image

Discussion

Our results suggest that SLR could be a useful parameter for discriminating between LGG and HGG. Using ROC analysis, we found an optimal threshold (2.88) with which to distinguish between HGG and LGG. It also showed a statistically significant correlation with the presence of free lipids and lipid-lactate peaks in the MRS sequences. We compared the SLR with the estimated concentration of lipid-lactate and lipid in MRS, a well-established method for grading glioma over the past two decades. The results showed a good positive and statistically significant correlation between these two modalities.

Energy production in the brain occurs solely through the glucose mechanism, and the presence of lactate is associated with impaired normal brain mechanisms [21]. In this case, the accumulation of lactate in the brain is due to O2 deficiency (ischemic changes), increased glycolysis [22], or necrosis [23], which is observed in all tumor grades. Elevated lactate concentrations can be seen in both benign and malignant tumors [23] but the combined lipid-lactate peak is associated with HGG [24]. While lipids are normally found in cell membranes and myelin sheaths, damage and necrosis in high-grade brain tumors cause them to convert to mobile lipids [25]. The distinct lipid peak in MRS is associated with glioblastoma (grade 4; [24]). The presence of these two peaks has been suggested in several studies to differentiate between low-grade and high-grade brain gliomas. In our study, we found a statistically significant difference between SLR in tumors and the presence of these two elements. In contrast to MRS, the IOP can localize the abnormality and receive signals from anatomically defined regions. This is an important difference in the brain because most brain abnormalities are focal and the layer of cranial fat that surrounds the brain contains significant amounts of the same compounds found in the abnormalities.

A study by Norlisah Ramli et al. [8] compared the use of chemical shift IOP sequences in 3‑T MRI with MRS for glioma grading. Their study included 22 patients, compared with 33 patients in our study; there were nine cases of LGG (all of them grade 2) and 13 of HGG. They showed a strong correlation between lipid concentration and SLR, and they expressed the SLR as median and interquartile range (IQR) values. Also, they found a significant SLR with advancing a grade. The optimal cut-off for discriminating between the two groups in their study was 0.064 versus 2.88 in our study, but they did not report the sensitivity and specificity of this value. The difference between these values in the study by Norlisah Ramli et al. and our study may be due to different magnetic fields, smaller sample size, and lack of sufficient grade 1 and grade 3 gliomas (only two patients). Pathological confirmation in their investigation derived from biopsies and could be affected by glioma heterogeneity and sampling error.

Another investigation conducted by De Pardieu et al. [18] showed that an SLR of > 9% and an SLR of > 20% could discriminate between grades 2–3 and grades 3–4, respectively. An AUC equal to 1 demonstrated 100% sensitivity and specificity for differentiating grade 2 and 3 gliomas in their study. Since the therapeutic approach is the same for grade 3 and 4 gliomas, we divided our patients into two groups (LGG and HGG). So far, our results could not be fully compared with their study due to this grouping approach, the SLR estimation formulas, and the use of MRI scanners with different magnetic fields. Furthermore, we also compared the measured fat content (by MRS) with the SLR and found a significant correlation.

Today, 3‑T MRI is routinely used for MRS. Our sample size was relatively small and we also did not examine the reproducibility of SLR measurements (interobserver agreement). We attempted to reduce the impact of glioma heterogeneity on our measurements by using three ROIs. These were limitations in our study. We believe that larger multicenter studies are needed to confirm our results. Employing 3D tumor segmentation in future studies for measuring sequences across the entire 3D tumor volume may offer more reliable statistical calculations with more data points, especially in calculations of in-phase opposed-phase sequences.

Conclusion

The signal loss ratio is a simple, rapid, and noninvasive marker for differentiating between low-grade and high-grade gliomas. There is a significant correlation with both the concentration and presence of free lipid and lipid-lactate peaks in magnetic resonance spectroscopy.