Magnetic resonance spectroscopy for enhanced multiparametric MRI characterization of [¹⁸F]FET PET-negative gliomas

Abstract

Background

Approximately 30–36% of gliomas presented with [¹⁸F]fluoroethyl-l-tyrosine ([¹⁸F]FET) PET-negative at primary diagnosis, which interferes with the differentiation of gliomas from other isolated brain lesions. Preoperative noninvasive identification of [¹⁸F]FET PET-negative gliomas to aggressive surgical treatment could reduce ineffective treatment and improve prognosis. This study aimed to assess the potential utility of multiparametric MRI with ¹H-magnetic resonance spectroscopy (¹H-MRS) in the diagnosis of gliomas within [¹⁸F]FET PET-negative isolated cerebral lesions.

Results

A total of 51 patients (mean age 44.35 ± 27.15 years, 26 males) with 37 gliomas and 14 non-gliomas were recruited for the study. More than half of PET-negative gliomas presented T2-FLAIR mismatch sign, whereas non-gliomas were more likely to present absence of T2-FLAIR mismatch sign (54.05% vs. 7.14%, p < 0.001). Choline to creatine (Cho/Cr) ratios in gliomas were significantly higher than those in non-gliomas (2.21 vs. 1.30, p < 0.001). Multiparametric MRI (AUC = 0.88) outperformed conventional MRI (AUC = 0.72) in differentiating gliomas from non-gliomas (NRI = 0.29, p = 0.02). And WHO grade was correlated with Cho/Cr and total lesion tracer standardized uptake (TLU) (r = 0.43 and 0.55; p = 0.007 and < 0.001; respectively). Low-grade PET-negative gliomas exhibit low levels of both TLU and Cho/Cr, but the distribution of TLU and Cho/Cr is more variable in high-grade gliomas. Furthermore, there was a moderated correlation between TLU and Cho/Cr in low-grade PET-negative gliomas (r = 0.54, p = 0.017), whereas there was no correlation in the high-grade PET-negative gliomas (r = -0.017, p = 0.95).

Conclusion

Multiparametric MRI with ¹H-MRS demonstrates significant promise in enhancing the diagnosis and overall clinical management for [¹⁸F]FET PET-negative gliomas. Moreover, the correlation between TLU and Cho/Cr that was affected by tumor grading of 2021 WHO criteria provides a rationale for further research into the mechanisms of reduced [¹⁸F]FET uptake in gliomas.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13550-025-01224-8.

Background

Preoperative non-invasive identification of gliomas has been a challenge and a crucial factor in glioma-related clinical treatment decisions. Positron emission tomography (PET) with radiolabeled amino acids including [¹⁸F]fluoroethyl-l-tyrosine ([¹⁸F]FET) tracer play an important role in localizing tumor lesions for biopsy, differential diagnosis, and recurrence determination due to the specific uptake of amino acid tracers by tumor cells [1–3]. [¹⁸F]FET PET imaging performed well in differentiating primary high-grade and low-grade gliomas (accuracy, 70.4–75%) [4, 5]. However, 30–36% of gliomas may interfere with the diagnosis due to [¹⁸F]FET PET-negative findings [2].

[¹⁸F]FET PET-negative cerebral lesions suspicious for gliomas mainly composed of demyelination, metastasis, lymphomas, and a small proportion of gliomas [2]. But the surgical resection is the necessity for clinical treatment of gliomas, which is distinct from the treatment of other lesions. And PET-negative gliomas performed a significantly prolonged progression free survival as compared to PET-positive glioma patients (23.1 ± 16.7 m vs. 16.4 ± 14.2 m; p = 0.003) [6]. Therefore, preoperative noninvasive identification of [¹⁸F]FET PET-negative gliomas to aggressive surgical treatment could reduce ineffective treatment and improve patient prognosis. However, inexperienced nuclear medicine doctors relying on PET imaging alone have difficulty definitively diagnosing [¹⁸F]FET PET-negative gliomas.

In contrast, multiparametric MRI integrating conventional and advanced MRI has been used for multidimensional assessment of gliomas preoperatively and postoperatively [7]. The T2-FLAIR mismatch sign based on conventional MRI was an imaging biomarker for the specific identification of IDH-mutant with 1p/19q non-codeleted (IDHmut-Noncodel) astrocytoma in lower grade gliomas, but its sensitivity was only 22–46% [8, 9]. However, the value of lesion location, enhancement, necrosis, and edema in distinguishing gliomas from other cerebral lesions remained contentious across various studies [10–12]. The utilization of advanced MRI, including magnetic resonance spectroscopy (MRS), diffusion-weighted imaging (DWI), and arterial spin labeling (ASL) imaging, facilitated the quantitative identification of gliomas in contrast to conventional MRI [13]. The minimal apparent diffusion coefficient (ADC_min) exhibited a significant elevation in tumefactive demyelinating lesions compared to high-grade gliomas [10]. MRS revealing the molecular composition of a specific tissue provided a distinctive understanding of the physiological or pathophysiological mechanisms underlying gliomas [14, 15]. Gliomas with enhanced infiltrative and invasive characteristics likely demonstrated increased choline to creatine (Cho/Cr) and choline to acetylaspartate (Cho/NAA) ratios [16, 17]. Malignant tumor lesions stimulate the neoangiogenesis in the brain, leading to an increase in cerebral blood perfusion (CBF), which forms the basis for utilizing ASL imaging to distinguish gliomas from other non-neoplastic lesions [18]. Moreover, several studies have demonstrated the higher precision of multiparametric MRI in identifying gliomas from other types of lesions than single parameter MRI [19, 20]. Therefore, we hypothesized that the application of multiparametric MRI could enhance diagnosing gliomas in [¹⁸F]FET PET-negative isolated cerebral lesions.

With the clinical application of hybrid PET/MR scanners in gliomas, exploring how to efficiently apply multimodal imaging to improve diagnostic accuracy is a research hotspot. Currently, some [¹⁸F]FET PET/MRI studies have focused on the preoperative diagnosis and tumor localization of gliomas [21–23]. The combination of [¹⁸F]FET PET and MRS could be used to differentiate glioblastoma from non-glioblastoma [24]. However, the additional and potential values of hybrid multiparametric MRI in the PET/MRI scanning for the preoperative diagnosis of gliomas in [¹⁸F]FET PET-negative isolated cerebral lesions remain not investigated. We aimed to underscore the potential value of multiparametric MRI in identifying gliomas with [¹⁸F]FET PET-negative findings to offer valuable surgical guidance and enhance the overall clinical management of these [¹⁸F]FET PET-negative challenging cases.

Methods

Study design and patients

Between March 2023 and May 2024, participants with isolated cerebral lesions who initially suspected gliomas based on conventional MRI at the Xuanwu hospital were prospectively enrolled. All patients underwent a brain multi-parametric [¹⁸F]FET PET/MR scan followed by biopsy or lesion resection surgery. Inclusion criteria included (1) age greater than 18 years, (2) no absolute contraindications to PET/MR examination, (3) conventional brain MRI initial suspicion of gliomas by one radiologist with more than 7 years of experience in the diagnosis of neuro-oncology and one neuro-oncologist with more than 20 years of experience, and (4) cerebral lesions showed a [¹⁸F]FET PET-negative presentation that the maximal tumor to background ratio (TBR_max) on PET images was less than 1.6 or absence of any visually increased signal abnormality in PET imaging according to PET imaging guidelines for gliomas [1, 2]. A participant was excluded if he or she met any of the subsequent criteria: (1) incomplete or poor-quality PET/MR images, (2) with malignant brain tumors of non-neuroepithelial origin diagnosed by pathology, (3) no surgical treatment or biopsy to acquire pathologic evidence, and (4) absence of 2021 World Health Organization (WHO) classification of tumors of the central nervous system taxonomy. Flow chart of inclusion and exclusion is shown in Fig. 1.

Fig. 1.

The flowchart of enrolled and excluded criteria. 189 patients suspected of gliomas were initially recruited to perform a brain [¹⁸F]FET PET/MR scanning. A total of 51 cases of [¹⁸F]FET PET-negative isolated cerebral lesions were enrolled after excluding 25 cases

Pathology

Pathology analysis of lesions was defined at surgery or biopsy performed by two pathologists with more than 10-year neuro-oncology pathological diagnosis in a blinded manner according to the 2021 WHO classification of brain tumors. Hematoxylin–eosin (H&E) and Ki-67 (ZSGB-BIO, Beijing, China, mouse monoclonal, diluted 1:50) immunohistochemistry with diaminobenzidine staining (Polink-1HRP Broad Spectrum DAB Detection Kit, Golden Bridge International, Mukilteo, USA) were used to initially evaluate the cell proliferative index of lesions for determining gliomas. The subsequent IDH mutation and chromosome 1p/19q co-deletion were identified by the whole-exome next-generation sequencing for classifying the glioma molecular type. Luxol fast blue (LFB) and H&E staining were performed to further validate demyelination for highly suspected demyelinating lesions. All images of section were captured with Leica Biosystems (Leica Aperio AT Turbo, USA).

PET/MR imaging

All [¹⁸F]FET PET/MR images were acquired with a 3.0T integrated PET/MRI (GE Healthcare) scanner using 19-channel head-neck coil. All patients adhered to a minimum fasting period of 4 h, followed by brain simultaneously MRI and PET scanning within a time frame of 20–40 min after the 185–200 MBq [¹⁸F]FET injection. The MR imaging sequences for the brain comprised 3D T1 weighted imaging (T1WI), axial T2 weighted imaging (T2WI), axial T2 fluid attenuated inversion recovery (T2-FLAIR), DWI, 3D ASL, 2D multivoxel ¹H-MRS, and an 3D T1 contrast enhanced (T1CE) sequence. MRS scans were localized at the maximum cross section of the lesion with enhancement on the T1CE images or absolute intensity maximum on T2-FLAIR images, and avoided contamination by normal cerebral tissue, the skull base, hemorrhage, and the ventricular system. And the MRS pre-scan data was required to achieve full width at half maximum less than 0.096 ppm and percentage standard deviation (SD) less than 20% before formal scanning. The PET images were acquired in 3D mode with an axial field-of view of 35 cm. An 18-s 2-point Dixon scan was acquired for MRI-based PET attenuation correction. And brain attenuation-corrected PET images were reconstructed by using the ordered subset expectation maximization algorithm (6 iterations, 16 subsets, and full width at half maximum of a Gaussian filter of 3.0 mm) under the time-of-flight technique with a 256 × 256 matrix and 2.78 mm slice thickness. Detailed scanning settings and parameters of PET/MRI sequences were summarized in Supplementary Table S1.

Imaging analysis

All PET/MR images were loaded into image processing workstation (AW4.7, GE Healthcare) to perform imaging analysis. ADC and CBF maps were respectively obtained from DWI and ASL imaging using workstation. Then PET, T2, T2-FLAIR, ADC, and CBF images were registered to the 3D T1 image with a 1 mm isovoxel, respectively. Since all cerebral lesions showed [¹⁸F]FET PET-negative (TBR < 1.6), the definition of volume of interests (VOI) of tumors was defined by two neuroradiologists manually contouring the abnormal high intensity area on T2-FLAIR or enhanced signal area on T1CE, avoiding necrosis, calcification, cysts, and hemorrhage (Supplementary Fig. 1) [1, 2, 25]. The maximum standardized uptake value (SUV_max), mean SUV (SUV_mean), TBR_max, TBR_mean, tumor volume (TV), and total lesion tracer standardized uptake (TLU = TV × SUV_mean) based on VOI were calculated for evaluating [¹⁸F]FET uptake.

Conventional MRI analysis consisted of lesion location, T2-FLAIR mismatch sign, hemorrhage, edema, necrosis, and enhancement. ADC_min and CBF_max were calculated from ADC and CBF images based on VOI. A circular (diameter 50 mm) was drawn in the normal cerebral hemisphere at the slice of the centrum semiovale including cortical and white matter for calculating the parameters of background cerebral tissue to acquire the relative ADC_min (rADC_min) and relative CBF_max (rCBF_max) [26]. Compound metabolism data of ¹H-MRS data were generated by workstation, providing automatic baseline correction, peak assignation, and ratio calculation. Metabolic measurements were performed by selecting 3–5 voxels to calculate the mean metabolic value from the voxels that were not significantly distorted baselines and located at the site with enhancement in T1CE or with high signal in T2 FLAIR images. The voxels containing normal cerebral tissue, the skull base, hemorrhage, and the ventricular system were excluded to calculate. The signal intensities were calculated by the sum of each peak, and relative ratios were calculated by the division of two concentrations. The noise level was defined as the variation of noise region (range 0.4–0.9 ppm). To avoid spuriously high-ratio values, voxels were excluded if the signal to noise ratio (SNR) of Cr was less than 3.In addition to recording the peak area ratios of Cho/Cr and Cho/NAA, the myo-inositol (MI) peaks and the lactate-lipid (LL) peaks were recorded in four levels (undetectable, positive and detectable, strongly positive, and extremely strongly positive) as described in pervious study [27].

Statistical analysis

The distribution of conventional and advanced MRI parameters between gliomas and non-gliomas group were analyzed by Chi-square test or Mann–Whitney U test/t-test. Furthermore, post-hoc multiple comparisons based on the Kruskal–Wallis test were used to compare differences in advanced MRI parameters among gliomas, demyelination, and metastasis. The MRI variables for which statistical significance existed for the above tests were continued to be used in univariate and multivariate logistic regression analyses to predict gliomas from all lesions. Receiver operating characteristic curve (ROC) and decision curve analysis (DCA) were used to evaluate the accuracy and clinical value of conventional and multiparametric MRI predicting gliomas. Besides, net reclassification improvement (NRI) was calculated to compare the difference in diagnostic efficacy for gliomas between conventional and multiparametric MRI. The correlation between MRI parameters and WHO grades was investigated by paired Spearman's tests.

All statistical analyses were conducted by using R software (version 4.3.1, Vienna, Austria) and python (version 3.7). A p value < 0.05 was regarded as statistically significant.

Results

Demographics of patients

Fifty-one patients (44.35 ± 27.15 years old, 26 males and 25 females) with [¹⁸F]FET PET-negative primary isolated cerebral lesions were prospectively included following excluding 25 cases (Fig. 1). There were 37 gliomas and 14 non-gliomas (8 demyelination cases and 6 brain metastases cases, Table 1). 19 PET-negative cases with WHO 1–2 grade gliomas were grouped as low-grade gliomas, and 18 PET-negative cases with WHO 3–4 grade gliomas were classified as high-grade gliomas (Table 1).

Table 1.

Demographics of enrolled patients

Variable	All patients (n = 51)
Age, years	44.35 ± 27.15
Sex
Female	25 (49.02%)
Male	26 (50.98%)
Pathology
Gliomas	37 (72.55%)
Non-gliomas	14 (27.45%)
2021 WHO grade
WHO grade 1	2 (5.41%)
WHO grade 2	17 (45.94%)
WHO grade 3	5 (13.51%)
WHO grade 4	13 (35.14%)
TBRmax	1.35 ± 0.28
SUVmax	1.80 ± 0.65

Data were described as mean ± SD or numbers (%). SD indicates standard deviation

Conventional MRI characteristics

Conventional MRI features distribution between gliomas group and non-gliomas group are summarized in the Table 2. There was a significant difference in proportion of T2-FLAIR mismatch sign between the gliomas and non-gliomas (p < 0.001). More than half of PET-negative gliomas (54.05%) presented T2-FLAIR mismatch sign, whereas non-gliomas were more likely to present absence of T2-FLAIR mismatch sign (94.44%). No significant differences of lesion location, enhancement, hemorrhage, edema, and necrosis were revealed between gliomas and non-gliomas (p = 0.806, 0.478, 0.753, 0.196, and 0.191, respectively; Table 2). Figure 2 illustrates the PET and conventional MRI findings of [¹⁸F]FET negative cases.

Table 2.

Conventional MRI parameters of gliomas and non-gliomas

Features	All types (n = 51)	Non-gliomas (n = 14)	Gliomas (n = 37)	χ2	p
Location				2.29	0.806*
Frontal	23 (45.10%)	6 (42.86%)	17 (45.95%)
Temporal	10 (19.61%)	2 (14.29%)	8 (21.62%)
Parietal	2 (3.92%)	0 (0.00%)	2 (5.41%)
Occipital	4 (7.84%)	1 (7.14%)	3 (8.11%)
Insular	7 (13.73%)	3 (21.43%)	4 (10.81%)
Others	5 (9.80%)	2 (14.29%)	3 (8.11%)
T2-FLAIR mismatch				9.23	0.003*
Absent	30 (58.82%)	13 (92.86%)	17 (45.95%)
Present	21 (41.18%)	1 (7.14%)	20 (54.05%)
Hemorrhage				0.53	0.478*
Absent	49 (96.08%)	13 (92.86%)	36 (97.30%)
Present	2 (3.92%)	1 (7.14%)	1 (2.70%)
Edema				0.10	0.753
Absent	20 (39.22%)	5 (35.71%)	15 (40.54%)
Present	31 (60.78%)	9 (64.29%)	22 (59.46%)
Necrosis				1.67	0.196
Absent	29 (56.86%)	10 (71.43%)	19 (51.35%)
Present	22 (43.14%)	4 (28.57%)	18 (48.65%)
Enhancement				3.31	0.191
Absent	24 (47.06%)	4 (28.57%)	20 (54.05%)
Rim	16 (31.37%)	5 (35.71%)	11 (29.73%)
Patchy	11 (21.57%)	5 (35.71%)	6 (16.22%)

Data were described as numbers (%). *Represented fisher's exact correction for chi-square test

Fig. 2.

PET and conventional MRI of [¹⁸F]FET PET-negative lesions. A A 55-year-old male with demyelination lesion. Conventional MRI showed patch enhancement isolated lesion located in the left central zone with negative [¹⁸F]FET uptake (TBR_max = 1.56). B A 46-year-old female with brain metastasis pulmonary adenocarcinoma proven by histopathology. Conventional MR images showed rim enhancement isolated lesion located in the left occipital lobe with negative [¹⁸F]FET uptake (TBR_max = 1.35). C A 29-year-old male with astrocytoma, WHO grade 2. Conventional MR images showed non enhancement isolated lesion located in the left insula lobe with negative [¹⁸F]FET uptake (TBR_max = 1.07). TBR_max indicates maximal tumor to background ratio

Advanced MRI parameters

The ADC_min (661.86 ± 199.63 vs. 598.50 ± 147.55 10^–6 mm²/s, p = 0.294) and rADC_min (1.15 ± 0.36 vs. 1.01 ± 0.35, p = 0.258) were not different between PET-negative gliomas and non-gliomas (Table 3). And no significant differences were observed in CBF_max, rCBF_max, CBF_mean, and rCBF_mean between gliomas and non-gliomas (all p > 0.05, Table 3). The Cho/NAA ratio (2.25 vs. 1.10, p < 0.001) and Cho/Cr ratio (2.21 vs. 1.30, p < 0.001) of gliomas were significantly higher than those of non-gliomas (Table 3). But there was no difference in MI and LL levels between gliomas and non-gliomas (Table 3). Moreover, multiple comparisons showed the Cho/NAA ratio of gliomas were higher than that of metastases (p = 0.007), but the difference in Cho/NAA ratio between gliomas and demyelination was not observed (p = 0.103; Fig. 3). And the Cho/Cr ratio of gliomas was higher than that of metastases and demyelination (p < 0.001, respectively; Fig. 3). The advanced MRI findings of [¹⁸F]FET negative cases are demonstrated in Fig. 4.

Table 3.

Advanced MRI parameters of gliomas and non-gliomas

Advanced MRI parameters	Non-gliomas (n = 14)	Gliomas (n = 37)	t/Z/χ2	p
ADCmin, 10–6 mm2/s	598.50 ± 147.55	661.86 ± 199.63	− 1.06	0.294
ADCmean, 10–6 mm2/s	1162.30 ± 283.35	1217.81 ± 217.46	− 0.73	0.469
rADCmin	1.01 ± 0.35	1.15 ± 0.36	− 1.14	0.258
rADCmean	1.29 ± 0.33	1.44 ± 0.29	− 1.56	0.125
CBFmax, ml/100 g/min	58.00 [53.00,96.00]	70.00 [58.00,86.00]	− 0.40	0.696
CBFmean, ml/100 g/min	29.22 [22.98,36.12]	38.09 [33.14,44.20]	− 1.90	0.059
rCBFmax	1.19 [0.92,1.53]	1.00 [0.81,1.24]	1.73	0.085
rCBFmean	0.92 [0.77,1.19]	0.89 [0.71,1.08]	1.16	0.250
Cho/NAA	1.10 [0.76,1.63]	2.25 [1.71,3.70]	− 3.32	< 0.001
Cho/Cr	1.30 [1.21,1.50]	2.21 [1.91,3.03]	− 4.00	< 0.001
MI			2.09	0.350
(−)	14 (30.43%)	32 (69.57%)
(+)	0 (0.00%)	4 (100.00%)
(++)	0 (0.00%)	1 (100.00%)
LL			5.23	0.155
(−)	9 (29.03%)	22 (70.97%)
(+)	5 (45.45%)	6 (54.55%)
(++)	0 (0.00%)	7 (100.00%)
(+++)	0 (0.00%)	2 (100.00%)

Dates were described as mean ± SD/median [IQR] or numbers (%). ADC indicates apparent diffusion coefficient; CBF, cerebral blood perfusion; Cho/Cr, choline to creatine; Cho/NAA, choline to acetylaspartate; LL, lactate-lipid; MI, myo-inositol; rADC, relative apparent diffusion coefficient; rCBF, relative cerebral blood perfusion

Fig. 3.

The distribution of Cho/NAA and Cho/Cr in the PET-negative lesions. A The pairwise comparisons of Cho/NAA showed that PET-negative gliomas had higher Cho/NAA than those of demyelination and metastasis. B The pairwise comparisons of Cho/Cr showed that both metastasis and gliomas had higher Cho/Cr than that of demyelination. Cho/Cr indicates choline to creatine; Cho/NAA, choline to acetylaspartate

Fig. 4.

Advanced MRI and pathology of [¹⁸F]FET PET-negative lesions. A A 55-year-old male with demyelination lesion. Advanced MRI showed lower diffusion (rADC_min = 0.92), mild-to-moderate increase of CBF (rCBF_max = 1.52), lower metabolism (Cho/Cr = 1.21 and Cho/NAA = 0.75). The demyelination with low Ki-67 expression was identified by the LFB and H&E staining (×200). B A 46-year-old female with brain metastasis derived from pulmonary adenocarcinoma. Advanced MRI showed mild diffusion (rADC_min = 1.19), lower CBF (rCBF_max = 0.72), lower metabolism (Cho/Cr = 1.29 and Cho/NAA = 0.67). The Ki-67 index was 0.08 (×400). C A 29-year-old male with astrocytoma, WHO grade 3. Advanced MRI showed mild diffusion (rADC_min = 1.10), lower CBF (rCBF_max = 0.84), markedly higher metabolism (Cho/Cr = 2.52 and Cho/NAA = 2.12). The tumor tissue had moderate proliferative activity (Ki-67 = 0.10, × 400). ADC_min indicates minimal apparent diffusion coefficient; CBF, cerebral blood perfusion; Cho/Cr, choline to creatine; Cho/NAA, choline to acetylaspartate; rCBF_max, relative maximum cerebral blood perfusion

Performance for identifying PET-negative gliomas

Both the univariate and multivariate logistic regression analyses revealed that T2-FLAIR mismatch and Cho/Cr were significantly associated with gliomas (p < 0.05). The Odds Ratio (OR) values and detailed p values of MRI parameters in logistic regression for predicting gliomas are described in the Table 4. Thus, nomogram integrating T2-FLAIR mismatch and Cho/Cr was established by logistic regression for predicting gliomas from PET-negative lesions (Fig. 5A). ROC analysis showed that AUC value of nomogram based on multiparametric MRI (AUC = 0.88) was greater than that of conventional MRI (AUC = 0.73; Fig. 5B). The accuracy of multiparametric MRI in differentiating gliomas and non-gliomas was 92%, with a sensitivity of 97% and a specificity of 79% (Table 5).

Table 4.

Univariate and multivariate analysis for predicting gliomas

Variables	Univariate analysis			Multivariate analysis
	OR	95%CI	p	OR	95%CI	p
T2-FLAIR mismatch	15.29	2.63–292.41	0.01	10.99	1.32–91.41	0.03
Cho/NAA	1.23	0.97–1.95	0.25	–	–	–
Cho/Cr	4.8	1.53–21.01	0.02	1.28	0.96–1.71	0.10

Cho/Cr indicates choline to creatine; Cho/NAA, choline to acetylaspartate; CI, confidence interval; OR, odds ratio

Fig. 5.

Nomogram and performance of multiparametric MRI for predicting gliomas. A Multi-lpparametric MRI nomogram integrating conventional MRI and MRS for predicting gliomas in PET-negative isolated cerebral lesions. B ROC curves of conventional and multi-parametric MRI nomogram. C DCA curves based on conventional and multi-parametric MRI nomogram. D Confusion matrix of multi-parametric MRI nomogram for predicting [¹⁸F]FET PET-negative gliomas. AUC, area under the receiver operating characteristic curve; DCA indicates decision curve analysis; ROC, receiver operating characteristic curve

Table 5.

ROC analysis for predicting gliomas

Model	AUC (95%CI)	Sensitivity	Specificity	Accuracy	Youden index
Conventional	0.73 (0.63–0.84)	0.54	0.93	0.65	0.47
Multi-parametric	0.88 (0.75–1.00)	0.97	0.79	0.92	0.76

AUC indicates area under the receiver operating characteristic curve; CI, confidence interval

Besides, the decision curve analysis indicated that the nomogram based on multiparametric MRI has a better net benefit of screening than the conventional MRI (Fig. 5C). The confusion matrix revealed that 4 cases were mistakenly classified by the multiparametric MRI nomogram in detecting PET-negative gliomas, as shown in the green on-diagonal entries of the confusion matrix (Fig. 5D). The NRI analysis showed a 29% improvement in the ability of multiparametric MRI to discriminate gliomas from non-gliomas in PET-negative lesions compared to conventional MRI (p = 0.02; Supplementary Table S2).

Advanced MRI characterization of PET-negative gliomas

Heatmap of Z-score showed that high-grade (WHO 3–4) PET-negative gliomas had higher Cho/NAA, Cho/Cr, and TLU than those of low-grade (WHO 1–2) gliomas (Fig. 6A). And WHO grade was correlated with Cho/Cr and TLU (r = 0.43 and 0.55; p = 0.007 and < 0.001; respectively; Fig. 6B). Simultaneously, there was a correlation between Cho/Cr and TLU (r = 0.33, p = 0.042), as shown in Fig. 6B. But distribution of TLU and Cho/Cr were different between high-grade gliomas and low-grade gliomas (Fig. 6C). Low-grade PET-negative gliomas exhibit low levels of both TLU and Cho/Cr, but the distribution of TLU and Cho/Cr is more variable in high-grade gliomas (Fig. 6C). Furthermore, there was a moderated correlation between TLU and Cho/Cr in low-grade PET-negative gliomas (r = 0.54, p = 0.017; Fig. 6D), whereas there was no correlation in the high-grade PET-negative gliomas (r = − 0.017, p = 0.95; Fig. 6E).

Fig. 6.

Correlation between advanced MRI parameters and WHO grades in PET-negative gliomas. A Heatmap of Z-score based on multiparametric MRI classified by WHO grade. B Heatmap of correlation between advanced MRI parameter and WHO grade. The WHO grade correlated with Cho/Cr and TLU (r = 0.43 and 0.55; p = 0.007 and < 0.001; respectively). And there was a correlation between Cho/Cr and TLU (r = 0.33, p = 0.042). Right skewed (red) ellipse represents positive correlation, and left sloping (blue) ellipse represents negative correlation. Blank squares represent no statistical correlation (p > 0.05) C Different distribution of TLU and Cho/Cr between high- and low-grade PET-negative gliomas. D Scatter plot showing moderated correlation between TLU and Cho/Cr in low-grade PET-negative gliomas (r = 0.54, p = 0.017). E Scatter plot showing no correlation between TLU and Cho/Cr in high-grade PET-negative gliomas (r = -0.017, p = 0.95). ADC indicates apparent diffusion coefficient; CBF, cerebral blood perfusion; Cho/Cr, choline to creatine; Cho/NAA, choline to acetylaspartate; SUV, standardized uptake value; TLU, total lesion tracer standardized uptake

Discussion

[¹⁸F]FET PET is a highly valuable clinical examination recognized by neurooncologists for the detection of gliomas. However, the identification of [¹⁸F]FET PET-negative gliomas represents a significant challenge for clinics, particularly in the situation where a glioma is suspected by conventional MRI. This study using an integrated PET/MR scanner analyzed the multiparametric MRI manifestations of common [¹⁸F]FET PET-negative isolated cerebral lesions and explored the combination of MRI parameters that can identify gliomas from other negative lesions. For the conventional MRI analysis, only the T2-FLAIR mismatch sign could distinguish gliomas from other PET-negative isolated cerebral lesions. Moreover, the Cho/NAA ratio and Cho/Cr ratio of gliomas were significantly higher than those of non-gliomas. Thus, a combination of T2-FLAIR mismatch and Cho/Cr performed better in identifying [¹⁸F]FET PET-negative gliomas. The correlation between Cho/Cr and TLU is noteworthy because it was different between high-grade and low-grade gliomas. This result has further strengthened our hypothesis that the metabolite of MRS is a potential tool for providing additional value for identifying [¹⁸F]FET PET-negative gliomas.

Differential distribution of T2-FLAIR mismatch between gliomas and non-glioams may be attributed to the high proportion of astrocytomas among the PET-negative gliomas, and other studies have confirmed that the T2-FLAIR mismatch sign could diagnose lower-grade astrocytomas with a specificity of over 90% [2, 9]. Previous studies have reported that astrocytomas constitute approximately 73–77% of [¹⁸F]FET PET-negative gliomas [6, 25, 28]. But in this study, the proportion of astrocytomas among PET-negative gliomas was found to be approximately 46% (17/37). The reason for this is that the 2021 WHO typing criteria were utilized instead of the 2016 WHO typing criteria in this research, resulting in the reclassification of some astrocytomas previously categorized based on the 2016 WHO criteria [29]. These reclassified astrocytomas are now identified as IDH wild-type glioblastomas (WHO grade 4) due to the lack of IDH mutations. Hiremath et al. [30] also found that the T2-FLAIR mismatch sign showed a statistically significant difference between demyelination and gliomas. Moreover, a recent search found that the T2-FLAIR mismatch sign in adult-type diffuse glioma lacking contrast enhancement demonstrated 100% specificity and 100% positive predictive value for the IDH mutation, and the absence of the T2-FLAIR mismatch sign (OR = 4.71, p = 0.008) was associated with 1p/19q codeletion [31]. This also indirectly supports our observations, which showed that 47.06% of PET-negative gliomas performed absence of contrast enhancement. Thus, the T2-FLAIR mismatch sign can help identify PET-negative gliomas and is a reliable conventional MRI biomarker for the diagnosis of PET-negative astrocytomas.

Some studies also found that other conventional MRI visual features could not discriminate gliomas from other non-glioma lesions [32, 33]. The reason for this is that gliomas, demyelination, and metastasis can present similar enhancement, edema, and necrosis in conventional MRI. Our study similarly demonstrated that lesion location, hemorrhage, edema, necrosis, and enhancement could not provide additional values for identifying [¹⁸F]FET PET-negative gliomas.

In accordance with the present results, previous studies have demonstrated that advanced MRI parameters could be used to differentiate gliomas from other non-glioma lesions [18–20, 34, 35]. Caravan et al. [34] found that ADC_min of high-grade gliomas were higher than those in brain metastasis. But in our study, the difference in ADC_min between gliomas and non-gliomas was not significant may be due to two factors. Firstly, our study focused on [¹⁸F]FET PET-negative gliomas, which exhibited lower tumor cell density activity compared to PET-positive gliomas. The results of some studies that confirmed that metabolic parameters of the amino acid PET are correlated with tumor cell proliferation activity in gliomas supported our hypothesis [36]. Secondly, the gliomas in this study included all grades of gliomas according to the 2021 WHO grading criteria, which is different from the previous studies that only focused on the high-grade gliomas. Furthermore, the correlation analysis confirmed that ADC values of PET-negative gliomas exhibiting lower tumor cell density activity were not correlated with WHO grades. Thus, the ADC parameters cannot be used for diagnosing PET-negative glioma grading or differentiating PET-negative gliomas from other lesions. Although ADC_min and rADC_min were not different between PET-negative gliomas and non-glioma lesions, the relative ADC values of PET-negative gliomas and non-glioma lesions were more than one. And numerous studies have similarly reported that low-grade gliomas exhibit rADC_min values greater than one, with astrocytomas in particular having higher rADC_min [12, 37, 38].

The ability of CBF parameters to discriminate gliomas from other cerebral lesions is controversial in many studies. Partial studies found that neoplastic lesions had a higher CBF than those of non-neoplastic lesions, and the CBF of glioblastomas was significantly higher than that of metastasis [18, 39]. Nevertheless, other studies demonstrated that non-neoplastic cerebral lesions may show increased perfusion in CBF map similar to high-grade gliomas [40]. We also found no statistically significant difference in CBF parameters between PET-negative gliomas and non-glioma lesions. A possible explanation for this might be that PET-negative gliomas have a lower cell density, resulting in a reduced requirement for cerebral blood flow supply. This could explain why the CBF parameters used in this study were unable to distinguish the gliomas from other cerebral lesions. And previous studies confirmed that CBF parameters of gliomas were associated with grades [41–43]. However, the correlation between CBF and grades was not observed in the PET-negative gliomas. This result may be explained by the fact that 2021 WHO grade of gliomas is more dependent on molecular markers. Therefore, the degree of amino acid uptake in PET-negative gliomas may not be correlated with the updated tumor grade. Vettermann et al. [25] also thought WHO grade may not affect the uptake of glioma [¹⁸F]FET mediated by LAT1, and they proposed the hypothesis that there are unknown biological processes affecting glioma [¹⁸F]FET uptake.

The single most striking marked observation to emerge from the data comparison was that the Cho/NAA and Cho/Cr of PET-negative gliomas were higher than those of non-gliomas, and Cho/Cr was an independent risk factor for identifying PET-negative gliomas in this study. These results were similar to those of Ikeguchiet al. [16] who also found that the Cho/NAA ratio was significantly higher in gliomas than in demyelination. Some studies also supported our findings that Cho/Cr of gliomas were higher than that of metastasis [44]. This result could be attributed to the stronger invasions of tumor cells in gliomas, which leads to a higher degree of local tumor cell infiltration compared to other cerebral tumors [17]. The Cho/Cr ratio indicates the metabolic activity of tumor cell membranes, which correlates with tumor proliferation and invasion. Consequently, within a single voxel, gliomas contain a greater proportion of indistinct margins (indicative of invasion) and tumor core, resulting in a higher Cho/Cr ratio compared to metastatic tumors and demyelinating lesions. And correlation analysis revealed a simultaneous correlation between Cho/Cr measured by MRS and TLU measured by PET, with WHO grades in PET-negative gliomas. Moreover, different correlation between TLU and Cho/Cr supported that MRS could identify PET-negative gliomas and provide potentially clues for subsequent studies on the mechanisms of low [¹⁸F]FET uptake in gliomas. In light of the previous finding by Vettermann et al. [25], it could conceivably be explained as that certain mechanisms associated with tumor grading or molecular typing have the potential to interfere with the transport of FET by LAT proteins of PET-negative gliomas while not affecting the changes of Cho/Cr in gliomas. Therefore, the multiparametric MRI nomogram that combined the T2-FLAIR mismatch sign and Cho/Cr performed best in identifying PET-negative gliomas.

We acknowledge that our research possesses three limitations. Firstly, constrained by the smaller sample size of PET-negative isolated cerebral lesions, the non-glioma lesions analyzed in this study consisted mainly of demyelination and brain metastasis. There is a need for future studies to explore the value of multi-parametric MRI for identifying PET-negative lymphomas, brain abscesses, and other cerebral lesions. Secondly, although an integrated PET/MR scanner was used to investigate the correlation between multiparametric MRI and PET-negative isolated cerebral lesions, it is necessary to use PET/CT and MR for further validation. Thirdly, the hotspot areas based on 2D MRS scanning may differ from the hotspot areas of ADC or CBF maps. This may have imparted a limitation to the ADC and CBF parameters, preventing them from offering supplementary diagnostic value when combined with MRS parameters. The 3D MRS scanning in the future has the potential to further improve the diagnostic utility of multi-parametric MRI.

Conclusions

An approach combining the T2-FLAIR mismatch sign and Cho/Cr can enhance the identification of [¹⁸F]FET PET-negative gliomas to better apply multiparametric MRI in the diagnosis and surgical decision of [¹⁸F]FET PET-negative gliomas. Moreover, the correlation between [¹⁸F]FET uptake and Cho/Cr affected by the new version of tumor grading related to molecular subtyping can support further research into the mechanisms of reduced [¹⁸F]FET uptake in gliomas.

Abbreviations

ADC_min

Minimal apparent diffusion coefficient

ASL

Arterial spin labeling

AUC

Area under the receiver operating characteristic curve

CBF

Cerebral blood perfusion

Cho/Cr

Choline to creatine

Cho/NAA

Choline to acetylaspartate

DCA

Decision curve analysis

DWI

Diffusion-weighted imaging

[¹⁸F]FET

[¹⁸F]fluoroethyl-l-tyrosine

IDHmut-Noncodel

IDH-mutant with 1p/19q non-codeleted

LL

Lactate-lipid

MI

Myo-inositol

MRS

Magnetic resonance spectroscopy

NRI

Net reclassification improvement

PET

Positron emission tomography

OR

Odds Ratio

rADC_min

Relative minimal apparent diffusion coefficient

rCBF_max

Relative maximum cerebral blood perfusion

ROC

Receiver operating characteristic curve

SD

Standard deviation

SUV_max

Maximum standardized uptake value

SUV_mean

Mean standardized uptake value

TBR_max

Maximal tumor to background ratio

TLU

Total lesion tracer standardized uptake

TV

Tumor volume

T1WI

T1 weighted imaging

T1CE

T1 contrast enhanced

T2-FLAIR

T2 fluid attenuated inversion recovery

T2WI

T2 weighted imaging

CNS tumors WHO classification

World Health Organization classification of tumors of the central nervous system taxonomy

Author contributions

JL, XRX, and XRL performed the conceptualization and study design. XRL, YC, XH, and BXC performed data curation, investigation, and formal analysis; MZ analyzed the pathology. HUL supplied the support of the methodology and visualization. JL provided resources, funding, and project administration. XRL and BXC performed the validation. XRL and XRX wrote the original draft. All authors have participated in critical revision and writing of the article. All authors have read and agreed to the published version of the manuscript.

Funding

This research has been supported in part by the National Key Research and Development Program of China (No. 2022YFC2406900), and the Huizhi Ascent Project of Xuanwu Hospital (HZ2021ZCLJ005).

Availability of data and materials

The datasets used during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was performed according to the guidelines of the Declaration of Helsinki and approved by the Xuanwu hospital ethics committee (protocol code: Xuanwu hospital [2023] 044). Written Informed consent was obtained from all patients involved in the study.

Consent for publication

The data presented have been approved for publication and each patient has provided written informed consent.

Competing interests

The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article.