In this study, we assessed the discriminatory potential of rCBV derived from DSC-PWI to non-invasively differentiate between IDH-mutant astrocytoma grade 4 and IDH-wildtype glioblastoma pre-surgically. Our voxel-wise approach, which accounted for volumetric segmentations and all percentile values, revealed that the most discriminative rCBV values lie within the lower percentiles of the non-enhancing regions. Here, though the values are overall low, they are notably higher in IDH-mutant tumors, suggesting the benefit of using an unsupervised rCBV selection approach over the conventional reliance on preselected mean or maximum values.

Furthermore, given the well-known coexistence of tumor infiltration and edema in the non-enhancing regions, we propose that these differential rCBV values may stem from varying degrees of tumor infiltration in these low-vascularized non-enhancing areas. Such regions may represent a greater degree of coexisting very-low vascularized infiltrated tissue in IDH-mutant cases, while in IDH-wildtype cases, they may more closely align with pure edema. Indeed, this hypothesis aligns well with prior knowledge: Glioblastomas are known to generate more pronounced edema, whereas Astrocytomas manifest a more substantial proportion of non-enhancing tumor tissue in the T2-FLAIR abnormality [25].

The observed elevated rCBV values in enhancing regions in both tumors, with no significant differences between them, would support the hypothesis that microvascular proliferation is a characteristic of grade 4 tumors, rather than a specific attribute of IDH-mutation status. Interestingly, the lower percentile rCBV values for Astrocytoma grade 4 tend to be slightly higher than those for Glioblastoma within these enhancing regions. This trend may suggest a higher homogeneity in Astrocytomas, characterized by a narrower range of rCBV values when compared to Glioblastomas.

Grade 4 Astrocytomas present morphological imaging traits that are distinct from grade 2–3 but are more reminiscent of IDH-wildtype Glioblastoma. The challenge of radiologically distinguishing between these two entities is highlighted by the morphological evaluations in this study, as illustrated in Fig. 2; Table 2, when considering the main markers described for differentiating between IDH-mutant and IDH-wildtype tumors [8,9,10,11,12,13].

Additionally, this study underscores the clinical significance of this differentiation in patients under 55 years old. In realistic clinical settings, the differentiation becomes crucial in this age group, making our findings especially pertinent. Unlike in those over 55 where the prevalence of the IDH mutation is negligible [1, 18, 19]. This approach mirrors a real-world clinical scenario where such differentiation is genuinely pertinent and impactful. For instance, the non-invasive presurgical differentiation of grade 4 astrocytic tumors is relevant beyond the ultimate histopathological diagnosis and could profoundly impact patient management across different levels. First, in specific scenarios, it could influence surgical decisions, such as whether to opt for function-preserving surgery or a biopsy (in cases of suspected grade 4 astrocytomas) versus total resection (in cases of suspected glioblastomas), particularly in challenging locations. Second, it may guide the sequence of the diagnostic workflow in histopathology and molecular pathology. For instance, by emphasizing and optimizing DNA sequencing utilization (often costly or difficult to access) in the most indicated cases to optimally detect IDH mutations. Ultimately, it offers an early prognosis prediction, which is invaluable, especially for young adults, and their families, enabling informed decisions and setting realistic expectations. Furthermore, such differentiation could be instrumental for the early detection of clinical trial candidates, for instance, for trials on treatments targeting IDH, which are anticipated to increase due to recent positive outcomes [2]. As we move further into the era of personalized and targeted therapies, the insights from our study could play an increasingly important role in shaping treatment strategies. This, in turn, hopefully will positively influence the disease course and enhance the quality of life for patients [3,4,5,6]. An illustrative example of potential clinical applicability of results in new patients with unknown diagnosis is shown in Fig. 5. Four additional illustrative cases are provided in Supplementary Material 4, along with the rCBVp30 values for the entire dataset.

Illustrative cases of two patients with unknown diagnosis: Patient_1 is 51 years old, and Patient_2 is 49 years old. The images display an extensive non-enhancing component beyond the enhancing tumor margins. This could be attributed to infiltrative tumor, edema, or a coexistence of both. rCBV color maps focused analysis allow the detection of small foci of slightly elevated rCBV (arrows) in the non-enhancing component of Patient_2, while it depicts clear areas of very low rCBV (arrows) in Patient_1. Quantification of the 30th percentile in non-enhancing areas indicates that Patient_1 has values that fall within the range of Glioblastoma (Gb), while Patient_2 aligns with Astrocytoma grade 4 (Astro 4). The diagnoses for both cases were histopathology confirmed

Several studies have attempted to identify IDH-mutation status using rCBV while analysing a range of adult diffuse gliomas. Some suggest the feasibility of discerning IDH mutation status, generally reporting higher rCBV values in both enhancing and non-enhancing regions for IDH-wildtype [17, 26, 27]. However, interpreting these findings requires caution, as these studies do not account for potential confounding with age or histological grade which are only reported as descriptive statistics, thereby preventing the optimal discernment of the specific differential in the current study. As an exemplification, considering that the vast majority of grade 4 astrocytic tumors are indeed glioblastomas, and the vast majority of grade 2–3 are IDH-mutant astrocytomas, a study claiming to identify IDH mutation status might actually be reflecting a more familiar differentiation between grade 2–3 and grade 4. Lastly, it is crucial to recognize that astrocytoma grade 4 is often either absent or significantly underrepresented in such studies, which limits the applicability of their results to this specific, smaller subgroup. This subgroup necessitates particular attention, as provided in our study.

Our literature search yielded only two DSC-PWI studies explicitly focused on grade 4 astrocytic tumors [28, 29], which in general terms reported higher rCBV values in IDH-wildtype tumors. However, due to different methodological approaches, direct comparison of results is not feasible. We consider relevant strengths of our methodology to include volumetric segmentations of easily demarcated morphological MR main tumor regions, which provide information on the entire abnormality; and the comprehensive evaluation of voxel-wise rCBV values through percentile analysis, not limited to preselected mean or maximum, which may obscure relevant differences in other parts of the full range of values.

Finally, another advanced MR technique deserving mention in this scenario is MR spectroscopy. It has been proven useful for IDH-mutation identification through specifically edited sequences, achieving high accuracies [30]. Also, more standard MR spectroscopy protocols offer information for glioma classification under the latest WHO guidelines [31]. However, the specific focused performance in grade 4 astrocytic tumors remains less clear because existing research again mixes tumor grades 2, 3, and 4. A potential limitation of this technique is its less extended implementation and use in neuroradiology departments worldwide compared to the widely extended and accepted DSC-PWI for brain tumor imaging [14, 15, 32, 33]. At any instance, recognizing the challenges, we believe that an ideal approach for the near future would combine comprehensive imaging data, including DSC-PWI and MR spectroscopy, with advanced data analysis techniques, such as AI and radiomics, to enhance presurgical tumor classification.

This study comes with several limitations. This is a single-site retrospective investigation. Nevertheless, this approach ensured data homogeneity, useful in pilot studies. The sample size, though seeming modest, is justified as all tumors were classified based on the stringent 2021 WHO Classification criteria, limiting retrospective patient inclusion. Also, IDH-mutant grade 4 astrocytomas are infrequent tumors, and they are rarely addressed in recent literature as a separate entity from their grade 2–3 counterparts. We recognize that theoretically, preloaded or low Flip-Angle (30º) DSC-PWI sequences might optimize rCBV measures when aligned with histological vascularization evaluations [14]. Yet, the primary differences lie in the non-enhancing region of tumors, where leakage-effects due to blood-brain-barrier disruption should be negligible, thus reducing the impact on rCBV calculations. Moreover, our study’s main focus wasn’t solely on this alignment. Different techniques have also shown reproducibility and robustness and we applied rigorous leakage correction procedures, mitigating potential leakage impacts [34]. Additionally, it should be highlighted that many clinicians have a preference for non-preloaded intermediate-high Flip-Angle sequences, particularly when it comes to the pre-surgical differential diagnosis [35,36,37,38,39,40,41]. This preference aligns with our study’s context and has demonstrated to be useful for diffuse gliomas’ genetic subtypes presurgical differentiation [42, 43]. However, our methodology can adapt to different DSC-PWI techniques with simple threshold adjustments [44]. Nevertheless, broader multicentric validations remain essential. Lastly, unfortunately, analysing a single case using our proposed methodology currently requires 10–15 min, hindered by the limitations of commercial software in PACS systems that force the use of multiple tools. This situation could impede rapid clinical adoption, but also highlights an opportunity for software enhancement in clinical neuroradiology, especially through improved segmentation tools and presenting CBV values via percentile analysis. With these improvements, post-processing time could potentially be reduced to around 2 min, underscoring the need for software advancements to narrow the gap between clinical practice and research in neuroradiology.

On the other hand, our study’s strengths are evident. All tumors were rigorously classified as per the 2021 WHO Classification criteria, ensuring contemporaneity. The tumor groups have been carefully balanced accounting for grade and age. Our insights hold clinical relevance from multiple discussed vantage points. Notably, the automatization of the data-extraction and data-selection ensures reproducibility minimizing operator-dependency. We underscore the importance of an unsupervised evaluation of the tumors’ entire percentile values, challenging the common clinical practice of relying on ROIs, mean or maximum values. In essence, our findings could be extrapolated to other clinical scenarios, laying the groundwork for further research.