Gliomas are the most common primary malignant central nervous system cancer, accounting for 80%-85% of malignant brain tumors (1), being highly aggressive and having the highest mortality and morbidity. The current standard of care treatment for glioma is maximal safe surgical resection followed by radiation therapy and concomitant chemotherapy (2). Despite the mainstream multimodal treatment, the median survival of most malignant glioblastoma patients has only improved to approximately 15 months (3), and less than 10% survive for over five years (4). According to the 2021 World Health Organization (WHO) criteria, gliomas are classified as grades I-IV (5), with different molecular subtypes and histopathology; these intrinsic subtypes showed different prognoses and outcomes (5). There exists a remarkable disparity in clinical treatment and prognosis between the different grades of glioma (3, 6). Thus, the preoperative differential diagnosis is essential for therapeutic decisions and determining the prognosis of patients with glioma.

MRI is the main workhorse for the evaluation of intracranial glioma (7). Modern diffusion MR protocols can acquire the diffusion data in a reasonable operating time and visualize the brain tissue in vivo at the micrometer scale. In a recent survey from Europe, beyond dynamic contrast enhanced (DSC)-MRI (67.6%), diffusion MRI (82.0%) was the most frequently used quantitative MR imaging technique in clinical neuroradiological practice (7). As yet, the apparent diffusion coefficient (ADC) from mono-exponential DWI, as an imaging biomarker, is the most frequently used diffusion parameter for tumor differentiation and monitoring of treatment efficacy (8). However, the mono-exponential model is susceptible to the microcirculation of blood in capillaries with inherent limitations in measuring the non-Gaussian distribution of water molecules. Le Bihan et al. (9) proposed the biexponential intravoxel incoherent motion (IVIM) DWI model with its aim to allow separation of molecular water diffusion from microcirculation, and to facilitate glioma differential diagnosis as well as identify the subtype of IDH mutation in gliomas (10). Moreover, there is evidence that the water molecular diffusion heterogeneity index (α) of the stretched exponential model (SEM) exhibits higher differential diagnosis sensitivity (92.9%) and specificity (100%) in grading glioma (11). Similarly, Zhang et al. also found that α showed the highest grading efficacy of glioma in their studies (12). Besides, there is evidence that diffusion distribution index (DDC), the other parameter from the SEM, is valuable in differentiating high-grade glioma (HGG) from low-grade glioma (LGG) compared to ADC (13). Chen et al. also suggested that DDC showed the highest sensitivity and specificity for glioma grading (14). Although the clinical values of various diffusion MRI models have been well proved in glioma differentiation by many researchers, few studies have explained the underlying histopathological significance under these parameters.

Tumor histopathology is the golden standard for the evaluation of glioma malignancy. According to the WHO guidelines, the following histological criteria were used to grade gliomas: cytological atypia, mitotic activity, cellularity, microvascular proliferation and/or necrosis (5). Mitotic activity is helpful in the distinction of grade II and grade III gliomas, whereas microvascular proliferation may suggest high-grade gliomas of grade III or grade IV (5). Some well-known ancillary markers of grading gliomas, such as anti-MIB-1/Ki-67 and anti-CD34, are closely related to tumor cell proliferation and microvascular density (5, 15). Anti-CD34 is a marker of vascular endothelial cells with high sensitivity and specificity of small and large vessels in both normal and tumor tissue (15). Furthermore, quite a few correlation studies have proved that DWI-related parameters are the potential non-invasive surrogates of histopathologic indexes, such as tumor cell density (16) and proliferative activity (17). Over the past decade of clinical research, IVIM-DWI derived parameters may simultaneously reflect tumor histology of microvessel density (MVD) or the percentage of MIB-1 expression positive rate(pMIB-1) both in humans and animals (18). Meanwhile, VERDICT MR with multiple b-value diffusion technique has been developed to reflect cell size (19, 20). However, although SEM-DWI derived parameters with the best differential performance (with highest AUC value of 0.968) in the differentiation of grades III from IV (14) for DDC and with the highest AUC of 0.993 in the differentiation of HGG and LGG (11) for heterogeneity index α, under which the exact correlation with histopathological biomarkers has rarely been reported (14).

However, due to not an exact match of tumor specimen and delineated tumor region of interest (ROI) on DWI images, prior studies’ correlation results have inherent limitations in identifying the DWI parameters as imaging biomarkers, which do not necessarily correspond to regions manifested on MRI images. Thus, the method of using the spatial location of each biopsy target based on a scanning coordinate system in an MR unit should be employed to improve the accuracy of the correlation study (21). This study correlated histopathological indexes with SEM-DWI parameters based on an intraoperative MR-navigated neurosurgical biopsy by a scanning coordinate system matching histopathological indexes and imaging parameters in gliomas.

Retrospective diagnostic study performed at one institution

This retrospective study was approved by the institutional Ethics Committee of our hospital.

In this study, SEM-derived intravoxel water diffusion heterogeneity α had a moderately to good negative correlation with CD34 expression rate in glioma by MDWI using a narrow-low range of multi-b-value combination. Meanwhile, we also observed a suggestive negative correlation between α and WHO grades. As such, α may serve as a potential indicator in identifying the malignant part of the intratumor by reversely indicating the CD34-stained vascularity. Meanwhile, SEM-derived DDC showed a significant negative correlation with pMIB-1, which may help non-invasively assess gliomas’ tumor proliferation.

As far as we know, many diffusion models, such as DKI, DTI, IVIM and FROC, have been analyzed for the correlations between diffusion parameters and pathological biomarkers in the prior studies on human gliomas. For example, DKI derived parameters correlated with the diverse nuclear-to-cytoplasmic ratio, which can be used to distinguish primary central nervous system lymphoma and high-grade glioma (23); and DTI parameters were ever discussed about their correlations with proliferation and microvascular density (24). IVIM biexponential parameters were recently discussed about the correlations between IVIM-derived fast diffusion and perfusion fraction parameters and VEGF- and MIB-1-positive rates in brain gliomas (25). However, despite of the usefulness of SEM parameters in grading gliomas (26, 27), they have not been analyzed on the correlations between the parameters and histological biomarkers in gliomas. Theoretically, the SEM derived heterogeneity index α is promising to indicate the intratumor’s water diffusion heterogeneity of gliomas according to its definition as an indicator of water diffusion heterogeneity index.

Several studies have suggested the particular usefulness of this SEM-derived intravoxel heterogeneity index α in grading gliomas (26, 27). Higher intravoxel diffusion heterogeneity with a lower α value in HGG than LGG is a well-known fact reported in prior studies (11, 28). Theoretically, the function of the stretched exponential model, also known as the Kohlrausch-Williams-Watts model (29), has been widely used in other fields of physics. From 2003 to 2004, this model was successfully introduced by Bennett and applied to the study of animal brain tissue and glioma model for nearly 20 years (30, 31). SEM-derived α enables the invasively quantitative evaluation of intravoxel heterogeneity in the distribution of diffusion coefficients for brain tissues and gliomas (30), which may not only identify normal brain cortex from white mater fiber in rat brain but distinguish the invading tumor cells in the peritumoral edema area surrounding the contrast-enhanced gliomas (30, 31). However, the exact biophysical implications of SEM-derived α remain to be elucidated in human brain gliomas, despite theoretically tortuous vasculature and heterogeneous cellular morphology presumed to be the leading causes of intravoxel heterogeneity to Bennett’s assumption (30). Through this practice-focused research, we confirmed our hypothesis that the performance of α have great association with clinical pathology, but still further studies are needed to confirm our results.

In this study, the heterogeneity index α was observed with a significant inverse correlation with anti-CD34 labelled vascularity in all gliomas. One possible explanation of such inverse correlation is that decreased α values indicates higher intravoxel diffusion heterogeneity, which is consistent with the characteristic feature of glioblastoma Anti-CD34 is a marker of vascular endothelial cells of small and large vessels in normal and tumor tissue (15, 32). Prior studies by Kong et al. (33) and Rahmah et al. (34) also suggested that CD34 overexpression is associated with higher grades of gliomas. It is well known that tumor areas with active vascular proliferation are often synchronized with tumor proliferation, and the primary tumor mass cannot grow beyond the size of about 1 - 2 mm without recruiting local host vasculature (22) or the formation of new vessels by angiogenesis (34). Meanwhile, to ensure the accuracy of pathological diagnosis, preoperative puncture targets determined by MR enhancement or T2FLAIR are often considered the most active tumor growth sites with an abundant tumor cell and vascular proliferation activity (21). Therefore, the abundance of CD34 - labelled microvessels represents the most malignant tumor area, which may be reflected using the heterogeneity index α. Therefore, as a reflection of the intravoxel heterogeneity in the distribution of diffusion coefficients for malignant glioma, α may not only serve as a useful imaging biomarker in helping glioma grading, but also a valuable imaging index of glioma heterogeneous assessment. Besides, there is evidence that tumor heterogeneity is a possible explanation of some observed drug-resistance (35), so α may also contribute to drug-resistance prediction.

Significant negative correlations were observed between DDC and pMIB-1 in grade II-IV gliomas regardless of the multi-b-value range (15 b values for 0-1500 s/m²) and 22 b values for 0-5000 s/m²) in this study. Previous researchers have ever performed a direct comparison of ADC values with the MIB-1(Ki-67), which demonstrated variable results from significant or moderately inverse correlation (28, 35) between ADC values and MIB-1 in brain gliomas. However, the mono-exponential model-derived ADC value, calculated by the assumption of voxels composed of one proton pool, was nonideal due to intratumor inherent heterogeneity when the number of contributing proton pools is unknown. Theoretically, DDC can be considered an ADC approximation weighted by the volume fraction of water in each part of the continuous distribution, and previous study conducted by Chen et al. found that DDC manifests stronger correlation with Ki-67 LI than the ADC derived from monoexponential model DWI (14). Previous studies (14, 36) have also performed the correlation analysis between DDC and pMIB-1. Moreover, to the best of our knowledge, most studies have ever performed the comparison between DDC and pMIB-1 by using the resected or biopsied samples, none of which strictly bridged their connection by point-to-point spatial coordinates in glioma patients. Due to the inherent intratumor heterogeneity in glioma (37), the resected tumor samples can’t be accurately matched with presurgical MR imaging parameters. The corresponding results varied widely and were hardly convincing in guiding the clinical practice. In this study, the center coordinates of the biopsy target were confirmed according to presurgical 3D-T1C (1mm in ST) or T2FLAIR (2mm in ST) images with prelabeled markers on the scalp for intraoperative MR navigation. Then, by the inbuilt software in the workstation, the coordinates of the selected biopsy were put into and labelled into the 3D-T1C or T2FLAIR navigation sequence to guide the following surgical biopsy. At last, on the SEM-derived DDC and α map fused with 3D-T1C or T2FLAIR, a square ROI with 90-120(mm²) in size, centered by the selected coordinates for biopsy, was placed on the tumor biopsy area of 3D-T1C or T2FLAIR. Compared to prior studies (38, 39), the matching of biopsied specimens and imaging ROI of the tumor was considered to a maximal extent. Meanwhile, the pMIB-1 was calculated using a multispectral phenotyping microscope and imaging analysis software, which ensured the accuracy of measurement to a maximum extent. Thus, the precise correlation analysis between DDC and pMIB-1 in glioma patients was of great significance in improving the understanding of the biophysical implications of DDC.

In this study, we chose DDC to correlate with pMIB-1 and reached consistent results with previous studies which used ADC as their variables. One possible explanation that could be attributed to such a phenomenon is that increased pMIB-1 indicates active cell replication and division, which indirectly reflects higher cellularity and malignancy, leading to the decrease of intercellular gaps and increase of diffuse coefficient. As a reflection of the distribution of diffusion coefficient, DDC calculated from the stretched-exponential model may also be affected.

Several limitations of this study should be considered in interpreting our findings. First, a multi-b-value DWI sequence based on 22 b values takes a longer scan time of 7 mins, which may lead to the failure of data acquisition due to the patients’ motion in clinical practice and requires to be improved in future work by reducing the number of b-values from 22 to 15 and lowering repetition time from 4000ms to 3000ms. Especially for those seriously ill patients, it was difficult for them to stay calm during the whole scanning process to obtain reliable data. Second, at the skull base, the measurements of DWI-MRI parameters on the tumor area may be inaccurate due to image distortion from inherent deficits of diffusion sequence. Although all the badly deformed cases and those with severe motion disturbance have been excluded in the present study, a slight distortion could not be completely avoided. Third, relatively small samples could bias the results, especially for the low-grade glioma samples, due to less frequency of clinical demand for surgical biopsy. Fourth, white matter tracts in the tumor contour were not reconstructed and analyzed, as the fiber tracts can significantly influence the diffusion parameters when the fibrous myelin sheath in the tumor remains intact, which should be considered in the future validation study. Last, although accurate matching method employed in the biopsy operation, we acknowledge minimal errors could produce from the typical three stages of ROI delineation, navigation registration error, and brain shift of biopsy operation, which is unavoidable under our current study design.

Our study indicates that SEM-derived parameters DDC and α are significant in histological grading of gliomas, DDC may indicate proliferative ability, and CD34-stained microvascular perfusion may be an important determinant of water diffusion inhomogeneity α in gliomas. Therefore, our results assert the promise of SEM-diffusion imaging for future clinical practice of glioma.

The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding authors.

The studies involving human participants were reviewed and approved by Huashan Hospital Ethics Committee. The patients/participants provided their written informed consent to participate in this study.

HP and YR contributed to the conception and design of the study. JuW wrote the first draft of the manuscript. YR and HL critically revised the manuscript. HZ performed the data acquisition and analysis. XD and WR performed the clinical assessment. HC, JuW, YZ, TQ, ZY and HL and revised sections of the manuscript. All authors contributed to manuscript revision and approved the submitted version.