Glioma is the most common primary tumor in the central nervous system. Clinically, glioma is usually divided into low-grade gliomas (LGGs) and high-grade gliomas (HGGs) based on the histopathological assessment. LGGs are well-differentiated, while HGGs are poorly differentiated and have a relatively poor prognosis (1–3). In recent years, more and more studies have shown that traditional histopathological grading of glioma has certain limitations due to the remarkable heterogeneity of tumors. For example, some LGGs overlap genetically with primary glioblastoma and show similar rapid disease progression (4, 5). It is difficult to distinguish them just by evaluating proliferation markers and cell morphology (6).

The 2016 World Health Organization (WHO) Classification of Tumors of the Central Nervous System officially listed molecular detection results as one of the important diagnostic bases for glioma classification for the first time (7), and the newly released 2021 guidance (WHO CNS5) emphasized the diagnostic value of molecular diagnosis for glioma subgroup (8). This substantial change has been achieved by further advancing the role of molecular diagnosis in the classification of CNS tumors, but still relying on other established methods for diagnosis of tumor characteristics including histology and immunohistochemistry. Isocitrate dehydrogenase (IDH) is a common molecular marker in glioma and is frequently used for predicting prognosis. Prior studies have shown that the prognosis of IDH-mutant gliomas is better than IDH-wild type gliomas (9, 10). This genetic grouping serves an important clinical indicator of stratifying tumors with differential susceptibility to adjuvant treatment. The biological similarities between some LGGs and glioblastomas make it critical to identify glioblastomas and separate them from more favorable IDH-mutant entities (11).

Unfortunately, preoperative distinction between different glioma grades and subtypes remains challenging with insufficient sensitivity and specificity. In addition, in the cases that gliomas at certain specific sites cannot be resected or punctured, or in patients who cannot undergo surgery due to age or other problems, the method based on image analysis can be used as a supplementary diagnostic tool for molecular classification of gliomas, and thus may have great potential value in treatment decisions (12). Magnetic resonance imaging (MRI) has been demonstrated a promising approach to non-invasively distinguish various tumor entities (13, 14).

MRI is also the preferred imaging method for glioma. Morphological characteristics and enhancement pattern of gliomas can be obtained by conventional MRI. Prior studies have shown that gliomas with different grades and IDH mutation status have differences in lesion properties such as location, internal signal and enhancement patterns (15, 16). Several imaging biomarkers contribute to the diagnosis of molecular subtypes of gliomas. Such as, T2-FLAIR mismatch (T2FM), which is a sign demonstrated a specificity of almost 100% for IDH mutant astrocytoma in recent studies (17, 18). However, morphological indicators are difficult to be quantified, so it cannot predict glioma grades and molecular subtypes accurately. Diffusion-weighted imaging (DWI) is an important sequence of MRI and serve for the identification and differential diagnosis in a broad spectrum of cancers (19). The assessment of cancers using DWI is based on the assumption that free water motion in tissues diminishes with growing tumor cellularity (20). The calculation of apparent diffusion coefficient (ADC) maps from DWI (at least with two b values) is a fast and straightforward procedure that can support grading and have shown the capability for IDH typing in gliomas (21–23). However, it is often difficult to identify the grades or even molecular subtypes of gliomas with single indicators obtained only by conventional MRI sequences. Few studies have combined MRI morphological features and ADC values to predict glioma grades and IDH mutation status. Additionally, consideration of patient age may help diagnosis because it has been shown that IDH-wild type gliomas are more common in older patients (11).

The purpose of our study was to explore the correlation between MRI morphological characteristics, ADC parameters and glioma grades, IDH mutation status. The developed multivariate predictive models may provide a new strategy for the formulation of glioma treatment, follow-up plan and prognosis evaluation.

The malignancy of glioma determines the choice of the surgical treatment plan and the prognosis of patients. The higher the tumor grade is, the worse the prognosis is. However, a single histopathological grade often has limitations. As in WHO grade IV glioblastoma, the degree of malignancy and prognosis may be different with different IDH gene types. WHO CNS5 in 2021 introduces a series of molecular diagnostic indicators on the basis of histological diagnosis, forming an integrated diagnosis and hierarchical reporting system, and defining multiple tumor types and subtypes (8). IDH gene family is still an important molecular marker of adult diffuse glioma. But something has changed. Previously, glioblastoma was diagnosed based on histological findings of microvascular proliferation and/or necrosis, including IDH mutations (10%) and IDH wild-type tumors (90%). In WHO CNS5, glioblastoma will contain only IDH wild-type tumors. IDH is a key rate-limiting enzyme of the tricarboxylic acid cycle (TCA), and IDH gene mutation in most gliomas occurs at the R132H site of IDH1 (24). Studies have shown that IDH mutation is an early event of glioma formation and has an important impact on glioma progression and tumor behavior (25). The clinical outcome of the IDH-mutant group is often better than that of the IDH-wild type group (26), and the IDH-wild type group is more aggressive, similar to the biological behavior of glioblastoma (17). Histopathological and immunohistochemical analysis is the final criteria for grade diagnosis and molecular subtype diagnosis of glioma, but there are the following limitations: (1) Internal heterogeneity and sampling bias of glioma may lead to errors in pathological results (27, 28). (2) The delayed diagnosis is not conducive to the formulation of surgical plan and the selection of preoperative treatment plan. (3) Some patients obtain pathological results by biopsy before surgery, but this is an invasive procedure, which may induce cerebral hemorrhage, epilepsy and other complications and increase the risk of iatrogenic injury (29, 30). MRI examination is an important auxiliary diagnostic method for glioma. MRI manifestations of glioma with different grades and IDH mutation status also have their characteristics. MRI can provide rich information for the diagnosis and prognosis evaluation of glioma. In this study, we successfully developed two noninvasive models by combing multiple new MRI features to distinguish low- and high-grade gliomas as well as with and without IDH mutation. These multivariate models led to a better predictive performance for glioma severity and IDH mutation than the single predictor.

In addition to IDH, many other molecular markers of glioma have been studied more and more in recent years, such as 1p/19q co-deletion, MGMT promoter methylation, TP53 mutation, EGFR amplification, etc. (31). These molecular markers have been confirmed to be related to the prognosis and treatment response in glioma patients. And in the fifth edition of the guidelines (WHO CNS5), glioblastoma, IDH-wildtype should be diagnosed in the setting of an IDH-wildtype diffuse and astrocytic glioma in adults if there is microvascular proliferation or necrosis or TERT promoter mutation or EGFR gene amplification or +7/-10 chromosome copy (8). Therefore, the importance of molecular markers for the diagnosis of glioma has once again attracted strong attention. However, it is difficult to predict the molecular subtypes of glioma with high accuracy by conventional MRI technology and general image post-processing simply. In recent years, classical machine learning approaches and deep learning approaches have shown the ability to identify the predictive features and to perform the actual prediction (32). Deep learning technology achieving performance that exceeds humans in the identification of content in images. So it can see the unseeable to predict molecular markers from MRI of brain gliomas. It is believed that a more reliable model can be used to better identify molecular markers of glioma through the combination of MRI and machine learning in the future.

Aging is usually associated with a poor prognosis of glioma. Consistent with the recent study, in our study, we found that high-grade glioma patients were older than patients with low-grade glioma (33). We further showed that age ≥ 60 years was an independent risk factor for predicting HGGs and IDH-wild type gliomas. A previous study reported that IDH-wild type gliomas are more common in elderly patients (34). We also found that the age of IDH-wild type gliomas was higher than that of IDH-mutant gliomas.

MRI morphological characteristics of gliomas with different grades and IDH mutation status may differ greatly. In this study, HGGs are more prone to cystic lesions and hemorrhage, which may be related to the high microvascular density and strong invasiveness of HGGs. We did not observe significant difference between IDH-mutant gliomas and IDH-wild type gliomas in terms of the cystic lesion, hemorrhage and peritumoral edema. However, Lasocki et al. showed that the proportion of edema was statistically significant between IDH-mutant gliomas and IDH-wild type gliomas. All five IDH-mutant patients had an edema rate of 5-33%, whereas most IDH-wild type patients had an edema rate of > 33% (44% of IDH-wild type patients had an edema rate of 34-67% and 14% had an edema rate of 68-95%) (35). The differences may be related to the subjective classification of the degree of edema and the selection bias of the enrolled cases.

The enhancement of glioma mainly depends on the degree of damage to the blood-brain barrier. The contrast agent retention in abnormal angiogenesis thus produces characteristic enhanced images (36). Therefore, MRI enhancement features are of high value for the judgment of the malignant degree of gliomas. In this study, most HGGs and IDH-wild type gliomas showed rim enhancement, while most LGGs and IDH-mutant gliomas showed hypovascular, which is basically consistent with previous literature reports (3, 11). Lasocki et al. suggested that unenhanced tumor volume > 33% was associated with IDH-mutant glioblastoma (35). Tumor location and distribution of lesions are one of the important factors affecting the prognosis of glioma patients. In this study, it was found that LGGs are more common to involve a single lobe, while HGGs are more common to involve multiple lobes, and more likely to involve corpus callosum and insula. The differences in the distribution of lesions in gliomas with different IDH mutation status have also been reported in previous literature. Nakae et al. showed that tumor location in the unilateral frontal lobe was highly correlated with IDH-mutant gliomas (P < 0.001) (37). Goze et al. found that 100% of LGGs with insula centers were IDH mutants (38). In our study, we also showed that compared with IDH-wild type gliomas, IDH-mutant gliomas were more likely to be associated with insula involvement, and frontal lobe involvement was more common in cases involving the single lobe.

A meta-analysis showed that ADC was significantly negatively correlated with the number of tumor cells in gliomas (39). The results of our study are consistent with previous studies, and the four ADC parameters of HGGs are lower than those of LGGs. The white matter cell substructure of HGGs was greatly damaged, and the diffusion of water molecules was limited. Therefore, In DWI imaging, HGGs showed higher signals, and LGGs showed lower signals. And accordingly, HGGs had lower ADC values (40, 41). In addition, The ADC value of IDH-wild type gliomas was significantly lower than that of IDH-mutant gliomas (13, 22, 23), which was consistent with the results obtained in this study. Patel et al. noted that histologically observed microcysts tended to increase in IDH mutant LGG cases, which could explain the higher ADC values in these cases, but the pathophysiological mechanism needed to be further elucidated (17).

In this study, we demonstrated the effectiveness of the combination of ADC parameters and MRI morphological features in predicting glioma grades and IDH molecular subtypes. The two multivariate logistic regression models combined with age, ADC parameters and MRI morphological qualitative parameters performed better in distinguishing the LGGs group from the HGGs group and the IDH-mutant group from the IDH-wild type group than using any single ADC value parameter alone.

However, this study also had some limitations, such as (1): This study was a single-center retrospective analysis, and the number of enrolled cases was relatively small. (2) Conventional MRI sequences and enhanced sequence were used instead of multimodality MRI imaging to obtain multi-parameter characteristics of tumors. Compared with other diffusion imaging, ADC has some limitations. (3) Subjective judgment errors may occur in the judgment of certain imaging features. For example, in some cases, it is difficult to distinguish simple vasogenic edema from non-enhanced tumors.

In conclusion, gliomas with different grades and IDH mutation status had significant differences in MRI morphology and ADC parameters. By combining age, MRI morphological characteristics and ADC value parameters, the accuracy of predicting histopathological grades and IDH molecular subtypes of glioma was greatly improved.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

All patients signed written informed consent before the enhanced MRI examination according to the regulations of Huadong Hospital, Fudan University. This retrospective study was exempted from ethical review and were conducted in accordance with the World Medical Association Declaration of Helsinki-Ethical Principles for Medical Research Involving Human Subjects.

SL and XF conceived and presented idea. ND, XZ and YY collected the data. RM, WS, LX, YS and GL analyzed the data. XX provided statistical guidance. ND drafted the manuscript. All authors reviewed the manuscript, and SL made corrections to the manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.