Diffusion-weighted imaging (DWI) is central to the diagnostic workup for patients with glioma (Barajas et al., 2016; Kolakshyapati et al., 2017; Zeng et al., 2018; White et al., 2019). A sign of poor prognosis is the presence of tumor hyperintensities on the DWIs, which are often referred to as regions of reduced diffusivity. The proposed underlying pathophysiological mechanism is a link between high tumor cellularity and reduced diffusivity (Chen et al., 2013; Laviolette et al., 2014). With stronger diffusion weightings (b = 2,000–3,000 s/mm²), the hyperintense regions are more conspicuous (Seo et al., 2008; Zeng et al., 2018), more useful for discriminating high- and low-grade gliomas (Seo et al., 2008), and better for prediction of the overall survival for patients with glioblastomas (Zeng et al., 2018). In a recent study, we observed that glioma conspicuity might be further improved by computing diffusion metrics based on so-called spherical tensor encoding (STE) (Nilsson et al., 2020), which at high b-values, yields an image contrast that is different from that obtained with conventional DWI (Szczepankiewicz et al., 2015).

Conventional DWI encodes for diffusion, along a single direction, at the time of the application of a pair of pulsed gradients (Stejskal and Tanner, 1965). Radiologic evaluation is performed by visual inspection of the DWI averaged over diffusion directions or by quantitative evaluation of the apparent diffusion coefficient (ADC) or its directional average—the mean diffusivity (MD). The MD is negatively correlated with tumor cellularity (Chenevert et al., 2000; Surov et al., 2017). The quantification of MD typically relies on DWI, with a moderate diffusion-weighting (b ≈ 1,000 s/mm²). When stronger diffusion-weightings are used (b ≈ 2,000–3,000 s/mm²), we may also estimate the mean kurtosis (MK) (Jensen et al., 2005), which is useful in glioma imaging because it enables their typing and grading (Tietze et al., 2015; Delgado et al., 2017).

Conventional DWI suffers from an inherent limitation, as it attenuates the MR signal in proportion to diffusion, but only along a single diffusion-encoding direction at a time. When the signal attenuation is low, we can only infer that diffusivity was reduced in that direction. However, distinctly different types of tissues can give rise to reduced diffusion. For example, a tightly packed dense tissue with high cellularity will exhibit low ADC, but so will the anisotropic structures when diffusion-encoding is applied along their short axis. If such anisotropic structures are not coherently aligned, we will find low ADC in many directions (Szczepankiewicz et al., 2016). Thus, the regions with high diffusion-weighted signal intensity in the directional average may reflect high anisotropy, as well as dense tissue with high cellularity. This problem arises because the conventional diffusion-weighted signal is sensitive to both isotropic diffusivity and diffusion anisotropy.

Tensor-valued diffusion encoding can be used to separate isotropic diffusivity and diffusion anisotropy by introducing a new measurement dimension—the b-tensor shape—that can be varied to obtain more information on the microstructure (Mitra, 1995; Mori and Van Zijl, 1995; Wong et al., 1995; Shemesh and Cohen, 2011; Eriksson et al., 2013; Lawrenz and Finsterbusch, 2013; Lasič et al., 2014; Szczepankiewicz et al., 2015; Henriques et al., 2021). Using the b-tensor terminology (Westin et al., 2016), a conventional DWI employs linear b-tensor encoding (LTE) as it encodes for diffusion along a single direction at a time. These images acquired with LTE are often orientationally averaged during post-processing (here referred to as LTE-DWI). By contrast, spherical b-tensor encoding (STE) sensitizes the signal to diffusion in all directions at a time (Lasič et al., 2014; Szczepankiewicz et al., 2016). The STE-DWI is, thus, already averaged across orientations during the acquisition. More importantly, however, is that STE attenuates anisotropic structures in proportion to their isotropic diffusivity so that a high signal is obtained only in regions with a low isotropic diffusivity, which is expected in tightly packed dense tissue with high cellularity. This means that STE-DWI is weighted for average isotropic diffusivity only but conventional DWI (LTE-DWI) for the combination of isotropic diffusivity and diffusion anisotropy. This effect is prominent at higher b-values. Conventional DWI entangles these two features even when it appears to be isotropic diffusion on the voxel scale. That is because this could be due to diffusion in isotropic structures or incoherent ordered anisotropic structures. Note that STE-DWI cannot be obtained from conventional LTE-DWI such as by the trace of the diffusion tensor imaging (DTI). The STE-DWI has previously been used to highlight the difference in cell density between the cortices of the cerebrum and cerebellum (Lundell et al., 2019; Tax et al., 2020; Vis et al., 2021) or for fast MD mapping (Mori and Van Zijl, 1995; Wong et al., 1995; Ianuş and Shemesh, 2018; Lasič et al., 2020; Szczepankiewicz et al., 2021a), but has not yet been systematically explored in gliomas.

The main hypothesis of this study is that the glioma tumors have higher conspicuity to the white matter at high b-value DWI obtained with STE (referred to as STE-DWI) than with high b-value DWI obtained with LTE and averaged across several directions (referred to as LTE-DWI). This is founded on the principle that gliomas exhibit reduced diffusion and low diffusion anisotropy because they are composed of densely-packed and approximately spherical cells, whereas the white matter is composed of elongated axons, which exhibit intermediate diffusivity and high anisotropy, as previously described (Beaulieu, 2002; Szczepankiewicz et al., 2015). Due to the orientation dispersion in white matter, we may never encode in a direction parallel with all axons in a voxel (Jeurissen et al., 2013). This causes LTE to yield less attenuation in white matter and reduces the contrast between white matter regions and regions with reduced isotropic diffusion. With STE, however, the MR signal of each part of the voxel is attenuated in proportion to its isotropic diffusivity. Accordingly, the white matter will appear darker (hypointense) with STE-DWI than LTE-DWI. Relative to the white matter, regions with reduced isotropic diffusivity will appear brighter (hyperintense) with STE-DWI (Szczepankiewicz et al., 2016). To test if gliomas are more conspicuous with STE-DWI than with LTE-DWI, we evaluated the contrast differences between LTE-DWI and STE-DWI both visually and quantitatively in a cohort of patients with glioma.

In this study, we compared the contrast mechanisms in glioma-related DWI hyperintensities obtained with STE-DWI and conventional LTE_DWI averaged across several directions at higher b-values (b = 2000 s/mm²). We found that STE-DWI improves the conspicuity of white matter-embedded tumor tissue compared with LTE-DWI. The finding was observed in multiple individual cases (Figure 1), and the potential diagnostic utility of STE-DWI was highlighted (Figures 4, 5). The improved conspicuity was also demonstrated quantitatively in Figure 3 using two quantities: the signal-intensity ratio—defined similarly to Seo et al. (2008)—and the contrast-to-noise ratio efficiency—applied similarly to Freitag et al. (2016). The former improved in all cases and in 70% of the cases. The improved conspicuity can be explained by opposite tendencies of glioma tumor tissue and white matter from the perspective of the signal attenuation (Figures 3A,B): STE-DWI attenuates anisotropic tissue components leaving only isotropic low-diffusion components contributing to the diffusion-weighed signal, which is prominent at higher b-values (Figure 2). Thus, STE-DWI emphasizes tissues, such as dense glioma tumors, with spherical and tightly packed cells that exhibit low diffusivity in all directions (Szczepankiewicz et al., 2015; Lundell et al., 2019; Tax et al., 2020; Vis et al., 2021). Signal attenuation in white matter was further illustrated in a patient without relevant MRI abnormalities (Figure 1, C1) and in a patient with a non-enhancing WHO IV glioblastoma without a DWI-hyperintensity (Figure 1, patient C2). Thus, STE-DWI attenuates healthy white matter to yield a contrast specific to hyperintensity.

STE-DWI may be relevant in the clinical settings for diagnosis or radiotherapy planning, where the presence of a pathology-related hyperintensity may improve detection or delineation. We expect the delineation and detection to be more accurate with STE-DWI because, unlike with LTE-DWI, a confounding factor (white matter) is suppressed as illustrated in Figure 5. The STE-DWI could even visualize information that is not accessible with conventional LTE-DWI. For example, tumor-related hyperintensities may exhibit similar signal intensity to white matter and a part of them is at risk of being missed or misinterpreted as a white matter if using only LTE-DWI (Figure 5, patient P5). Furthermore, STE-DWI could visualize tumor-related hyperintensities within white matter tracts (Figure 5, patient P6), which is a route of spreading for glioma tumor cells (Giese et al., 1996; Giovanna and Kaye, 2007). Including STE-DWI may reveal hyperintensities that, otherwise, could be missed (Figure 4, patient P4). Finally, the white matter that is shifted by adjacent tumor lesions is not depicted on STE-DWI, but only on LTE-DWI (Figure 1, patient C1), which could corroborate the differential diagnosis. The highest signal differences between LTE-DWI and STE-DWI were observed predominately in major white matter tracts (Figure 6). Hence, we expect STE-DWI to better show lesions “hidden” under white matter in these tracts and these anatomical locations could be of interest for applications of STE-DWI.

Conventional DWI at high b-values (3,000 s/mm²) has been shown to improve glioma conspicuity, compared with DWI at regular b-values (1,000 s/mm²) (Seo et al., 2008; Zeng et al., 2018). The analysis methods based on high b-values acquisitions, such as restricted spectrum imaging, may also increase glioma conspicuity (White et al., 2013). Another method of white matter suppression is based on subtraction the axial diffusivity (Freitag et al., 2016). It has shown to be useful in white matter tracts that have a coherent fiber direction but may not, unlike STE-DWI, suppress the signal adequately in fiber crossing regions. Tractography can also be applied to delineate glioma tumors from white matter (Nilsson et al., 2018a). However, the tractography based on LTE data only had limited accuracy (Maier-Hein et al., 2017) and might be improved by also including STE-data (Cottaar et al., 2020; Jeurissen and Szczepankiewicz, 2021). Other advanced techniques for improving DWI of gliomas include, for example, neurite orientation dispersion and density imaging (NODDI), which has been applied to differentiate glioma grade (Maximov et al., 2017; Li et al., 2019; Kadota et al., 2020) or to differentiate glioma from solitary brain metastasis together with mean apparent propagator MRI (MAP-MRI) (Mao et al., 2020). Microstructure models, such as VERDICT, have also been applied to map glioma microstructure (Zaccagna et al., 2019). Only a limited number of papers have based their investigation on data acquired with both LTE and STE (Nilsson et al., 2020; Li et al., 2021). What is different with the present approach is that it focuses on the contrast of STE-DWI itself, compared with LTE-DWI without modeling. It is, therefore, easy to integrate into a radiological workflow.

The STE-DWI has several advantages compared with LTE-DWI. First, STE-DWI produces images weighted by the isotropic diffusivity, whereas conventional LTE-DWI produces images weighted for the combination of isotropic diffusivity and diffusion anisotropy. This increases the specificity of the contrast. Note that STE-DWI is acquired using custom gradient waveforms for the encoding (Szczepankiewicz et al., 2021b) and, thus, cannot be obtained by post-processing of conventional LTE-DWI, such as by the trace of the diffusion tensor imaging (DTI). Second, STE-DWI is designed to be independent of the orientation of the object (rotationally invariant) that is already in the data acquisition step (Szczepankiewicz et al., 2021b), which also allows for greater flexibility in the design of imaging protocols (Nilsson et al., 2020). By contrast, in LTE-DWI, the signal must be acquired across sufficient directions to make it rotationally invariant (Mori and Van Zijl, 1995; Wong et al., 1995; Szczepankiewicz et al., 2019). Third, STE-DWI is feasible at common clinical platforms (Szczepankiewicz et al., 2019; Nilsson et al., 2020) and the scan times are generally short: we acquired a 6-average high b-value STE-DWI in just 30 seconds at an isotropic resolution of 2.3 mm³ (Figure 7). There are, however, also drawbacks. The major ones are that STE has higher demands on the gradient system and that it requires longer echo times (TE) and, thus, provides comparatively lower SNR. In particular, an encoding strength of b = 2,000 s/mm² can be achieved in approximately 15 ms shorter echo time by using a stand-alone optimized conventional pulsed-gradient LTE-DWI sequence (Stejskal and Tanner, 1965), when compared to STE-DWI on a gradient system with maximal gradient amplitude of 80 m T/m and slew rate of 100 T/m/s (Szczepankiewicz et al., 2021b). In this study, we acquired a TE of 84 ms for both STE and LTE, which is close to the currently achievable minimal TE for the STE at b = 2,000 s/mm² and this hardware configuration. This corresponds to an SNR reduction for STE-DWI of around 20%, under the assumption of exponential T2 signal decay in white matter with a T2 relaxation time of 80 ms (Whittall et al., 2002). The CNR_eff of LTE-DWI would, thus, increase by 20%. Lesions that are also hyperintense on T2w FLAIR images have longer T2-relaxation times and suffer less from this penalty. How the SIR would change for an optimized LTE-DWI sequence is harder to predict, as it depends on the relative difference in T2 between the hyperintensity and the white matter. Within age-associated white matter lesions, there is a substantial difference between the T2-relaxation time within and outside of axons, which leads to a considerable dependence of the diffusion-weighted signal on the TE (Lampinen et al., 2019). We expect this to also be the case for tumor-related hyperintensities in T2w images but are not aware of any detailed analysis. Note that diffusion-relaxometry techniques could be of interest here to add microstructure specificity (De Almeida Martins and Topgaard, 2018; Slator et al., 2021; Narvaez et al., 2022). This was, however, outside the aim of the present work, which was to study the contrast mechanism provided by STE-DWI without modeling or optimizing the imaging protocol.

The histological underpinnings of DWI- and T2/FLAIR-hyperintensities remain elusive, although it has been proposed that the pathophysiological mechanism of the reduced diffusivity in DWI-hyperintensities can be high the tumor cellularity (Chen et al., 2013; Laviolette et al., 2014). However, these may also appear to be due to prolonged T2 relaxation times in the tumor (T2-shine through effect). Even though the T2/FLAIR-hyperintensities, which are often in the tumor regions on MRI scans, coincide with DWI-hyperintensities, their pathophysiological mechanism at the tissue level in the tumor lesion could be different. It is proposed that these are associated with irreversible myelin or axonal loss, enlarged perivascular spaces due to lack of drainage of interstitial fluid, increased protein concentration that could lead to edema, or due to gliosis (Melhem et al., 1997; Ho et al., 2012; Haller et al., 2013).

We identified five limitations of the present study. First, the study lacks histological assessment of the tissue composition in the DWI-hyperintensities, and, thus, we cannot verify its histological correlate. Thus, the clinical usefulness of STE-DWI is only indicative at this point. Future studies, including comparisons with histological findings or targeted biopsies, would be highly beneficial. The second limitation is that we did not compare the conspicuity with other imaging modalities, such as perfusion MRI, or with metrics that can be extracted from the data by modeling, especially those related to tissue anisotropy. The third limitation is related to the acquisition protocols. The LTE-DWI was averaged over only 6 directions, which could lead to some residual rotational variance. However, the rotational variance is below 1% in structures with FA below 0.6 (Szczepankiewicz et al., 2019), and glioma tumors generally have low FA. For the STE-DWI, gradient waveforms were not compensated for concomitant gradient effects. These may introduce a position-dependent bias and future studies should use waveforms with minimized concomitant gradient effects (Szczepankiewicz et al., 2021b). Fourth, the contrast could also depend on the TE. The aim of the study was, however, to study the influence of the b-tensor shape on the image contrast. The fifth and last limitation is the limited sample size. Larger cohorts are needed to assess the improvements in tumor conspicuity in clinical settings.

We present initial evidence that STE-DWI at high b-values enables a more specific assessment of tumor hyperintensities in glioma tumors by suppressing the confounding contribution from white matter.

The data analysis were processed by a software package for diffusion MRI available at: https://github.com/markus-nilsson/md-dmri (Nilsson et al., 2018b). Project-specific code, MRI protocol and waveforms are available at: https://github.com/jan-brabec/tensor_valued_gliomas_in_vivo.

The project was approved by the local ethics committee-the Regional Ethical Review Board in Lund (2016/531, 2017/866 and 2018/993), and written informed consent was obtained from each volunteer according to recommendations of the declaration of Helsinki.

JB: conceptualization, investigation, formal analysis, writing - original draft, methodology, and writing - review and editing. FD: data curation, resources, and writing - review and editing. FS: conceptualization, data acquisition, investigation, software, methodology, and writing - review and editing. PB and BL: investigation writing - review and editing. AR: resources writing - review and editing. LK: funding acquisition writing - review and editing. C-FW: funding acquisition and writing - review and editing. PS: conceptualization, supervision, visualization, project administration, resources, methodology, funding acquisition, and writing - review and editing. MN: conceptualization, supervision, project administration, investigation, writing - original draft, writing - review and editing, and software, methodology. All authors have read and approved the final version of the manuscript.

MN declares ownership interests in Random Walk Imaging, and patent applications in Sweden (1250453-6 and 1250452-8), in the United States (61/642 594 and 61/642 589), and via the Patent Cooperation Treaty (SE2013/050492 and SE2013/050493). JB, MN, FS, and PS are inventors on pending patents pertaining to the methods presented herein in the United States (63/293 098). The remaining 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.