Differentiating IDH-mutant astrocytomas and 1p19q-codeleted oligodendrogliomas using DSC-PWI: high performance through cerebral blood volume and percentage of signal recovery percentiles

The imaging-based presurgical differentiation between astrocytomas and oligodendrogliomas represents an unsolved challenge in neuro-oncology. This differentiation has potential implications for diagnosis and initial patient management in several ways. First, accurate radiological classification allows for the selection of the most time- and cost-effective diagnostic test sequence in histopathology and molecular pathology [1‐5]. In certain situations, it can further enhance the accuracy of the final diagnostic process. For instance, some researchers advocate for additional molecular testing for 1p19q co-deletion when there is a discrepancy between neuroimaging and fluorescence in situ hybridization results [5]. Secondly, the differentiation between these two entities assists in choosing the most appropriate surgical approach (e.g., total resection vs. partial resection or biopsies), predicting prognosis, and deciding the necessity and intensity of different adjuvant therapies, all of which can vary between astrocytomas and oligodendrogliomas [1‐4].

Despite significant advances in imaging techniques, achieving accurate differentiation preoperatively remains complex. This emphasizes the need for innovative approaches that can enhance current diagnostic capabilities [6]. Qualitative morphological imaging, particularly through the T2-FLAIR mismatch sign, provides valuable insights and is highly specific for identifying IDH-mutant astrocytomas [7]. However, this method has low sensitivity, and its interpretation remains largely visual and subjective, reliant on the experience of the neuroradiologist with possible uncertain cases, especially for those less experienced. As a result, there has been an increasing interest in quantitative sequences. These offer the potential for more objective and reproducible results, enhancing the precision and reliability of presurgical tumor differentiation [8].

Dynamic susceptibility contrast perfusion-weighted imaging (DSC-PWI) is one such quantitative sequences that has received extensive attention in neuro-oncology. Widely available and included in the most recent brain tumor imaging consensus, DSC-PWI monitors T2* signal changes dynamically during the vascular passage of a gadolinium-based contrast agent (GBCA), thereby generating time-intensity curves. The standard evaluation involves calculating the area under the curves to obtain cerebral blood volume (CBV), which estimates the overall vascularization of the tumors [9‐11]. However, the potential information embedded in these time-intensity curves goes beyond just CBV. For instance, the percentage of signal recovery (PSR) is another metric that has received significant attention and demonstrated promising results for differential diagnosis of brain tumors [12‐14].

The PSR provides an indirect measure of T1 and T2* leakage effects, which notably influence the time-intensity curve after the first pass of GBCA. It has been extensively studied in tumors with well-known prominent blood-brain barrier (BBB) disruption like glioblastoma or lymphoma [13‐20]. However, in the context of a more preserved BBB, such as in grade 2–3 gliomas [21], PSR biological meaning may be different and it is scarcely investigated.

Commonly, CBV (and PSR) calculations focus on mean values from two-dimensional (2D) regions of interest (ROIs), or whole-tumor three-dimensional (3D) volumes of interest (VOIs). Some authors suggest using extreme (maximum/minimum) values, either automatically generated or manually identified via hot-spots of CBV. Nevertheless, some of these methods have their shortcomings. First, 2D ROIs only evaluate isolated regions of the whole tumor and overlook tumor heterogeneity [12]. Second, relying on preselected single values, such as the mean or maximum, may overlook potential significant differences within the entire spectrum of voxel-wise values. These values can be further investigated using histogram-derived percentiles [22].

Finally, when CBV and PSR are evaluated, they are treated separately and their performances are compared as if one must choose between them. However, the reality is that they offer distinct information that characterizes tumor vascular-microvascular habitats from different perspectives. Therefore, CBV and PSR could be integrated, providing additive rather than mutually exclusive information [23].

Taking all these considerations together, the authors’ objectives in the context of pre-surgical DSC-PWI evaluation of astrocytomas and oligodendrogliomas are:

This retrospective study was approved by the Research Ethics Committee of Hospital Universitari de Bellvitge.

In this study, we delved into the distinctive features of DSC-PWI in distinguishing astrocytomas and oligodendrogliomas. Firstly, we have pioneered the exploration of the PSR as a potential discriminator between these tumor types. Our data indicates a lower PSR in oligodendrogliomas. Regarding nrCBV, our findings align with previous studies that observed higher nrCBV in this tumor type [32‐36]. Furthermore, we have augmented the efficacy of PSR and nrCBV analysis by introducing an unsupervised method to identify the most discriminative voxel-wise percentiles (AUC-ROC PSRp70 = 0.84 and nrCBVp75 = 0.80), which exceeds the conventional approach of solely relying on single preselected values such as the mean or maximum. Notably, the combined use of nrCBV and PSR does not result in exclusivity, but rather in an additive effect, thereby forming a powerful basis for robust classification. An illustrative example of the applicability of our methodology to classify future presurgical patients is presented in Figs. 4 and 5.

The most common approach for nrCBV calculations involves focusing on the mean values of 2D ROIs manually depicted, or whole-tumor 3D VOIs. However, if a specific ROI is manually selected, operator dependency resurfaces, complicating reproducibility, and only a portion of the whole tumor is evaluated. Also, if only preselected values of the entire 3D tumor are used, the heterogeneity of the tumors (especially of oligodendrogliomas in the current differential) is overlooked. Indeed, relying on preselected values often assumes that they will capture the most significant differences, an assumption that may lack a robust foundation [12]. Therefore, these methods fail to consider that differences may exist anywhere across the full spectrum of voxel-wise values [22]. In this work, we propose an objective, unsupervised, voxel-level percentile evaluation of 3D segmentations of the whole tumor. This offers more reproducible, robust, and improved information that better captures tumor heterogeneity while avoiding weak assumptions in the preselection of optimal values to differentiate between entities.

In terms of novel insights about the PSR, to our knowledge, only two prior studies involved PSR evaluations in tumors with oligodendrocytic component. However, these studies were conducted prior to the establishment of the current 2021 WHO classification criteria and treated oligodendrocytic tumors as mixed with other low- and high-grade glial tumors [37, 38]. Therefore, their results are not applicable to the current classification system and are not comparable with ours. PSR uniquely provides indirect measures of the interplay between T1 and T2* effects. In general terms, lower PSR values indicate predominant T2* effects, while higher values suggest predominant T1 effects. Based on our current understanding of leakage effects, it is known that PSR is shaped by a complex interplay of BBB disruption, variations in cell size and density, and the size and tortuosity of vessels. This theory mainly applies to brain tumors with extensive BBB disruptions. These disruptions enable a portion of the GBCA to leak through the vessels and interact with the extravascular extracellular space (EES) molecules and adjacent cellularity, while another portion remains within the vessels, reflecting vascular features [39‐41]. However, when dealing with brain tumors with a more preserved BBB, such as the entities under investigation, the GBCA does not so significantly leak, thus minimizing the influence of BBB permeability and cellularity on the PSR. Consequently, it is plausible that the PSR in these tumors, such as grade 2–3 gliomas, primarily offers insight into the vascular architecture [42]. In simplified terms for clarity, in such cases, increased vessel disorganization and tortuosity could partially retain intravascular gadolinium. This incomplete wash-out could prevent the signal intensity curve from fully recovering, leading to a lower PSR. It is noteworthy that PSR calculation is not universally standardized. However, this specific vascular characteristic may be enhanced when the metric is calculated early towards the end of the first vascular pass of contrast, as done in this study. At this point, any potential GBCA leakage due to eventual foci of partial BBB disruptions is expected to be minimized.

This theory is further supported by previous observations in CNS tumors that lack a BBB and exhibit prominent tortuous vasculature, such as hypervascular metastases, meningiomas, or hemangioblastomas. These tumors are well-known for presenting with low PSR, attributable to their distinctive vascular features. Despite the different underlying explanation, these tumors share one commonality with the tumors currently under investigation: the absence, for one reason or another, of BBB leakage [22, 23, 26]. At this point, in the case of astrocytomas and oligodendrogliomas, it is crucial to point out that the histological vascular characteristics of oligodendrogliomas are marked by a network of irregular and tortuous vessels, frequently described as a chicken-wire capillary pattern [43‐45]. This vascular pattern diverges significantly from that found in astrocytomas which while also altered generally retains a greater similarity to normal brain tissue [46]. This knowledge, especially about the unique microvascular architecture of oligodendrogliomas, aligns with our observations.

On the other hand, when nrCBV and PSR are evaluated, they are generally treated separately and their performances are compared as if one must choose between them. However, the reality is that they offer distinct information that characterizes tumor vascular features from different perspectives. In fact, and especially if correction methods are applied to nrCBV, PSR should not influence nrCBV values and vice versa, as also evidenced by our correlation analysis. Given these considerations, nrCBV and PSR could be combined, providing additive rather than mutually exclusive information [12, 23]. Additionally, by considering both metrics, we are considering the tumors’ vascular-microvascular habitats from a broader perspective than what single variables allow. Specifically, our combined nrCBV and PSR classifier demonstrates promising potential, with an estimated AUC-ROC of 0.87, an excellent result which still requires further validation. Beyond improved classification results, the simultaneous consideration of multiple DSC-PWI variables offers a more comprehensive perspective of the tumor’s vascular characteristics than traditional single metrics do. Therefore, this approach is likely to be more robust, mitigating single metric variabilities and providing a more holistic assessment of the vascular characteristics of the tumors.

Additionally, our T2-FLAIR mismatch subanalysis suggests that our method retains good to excellent performance in those tumors where the T2-FLAIR mismatch is absent or uncertain (such as in scenarios involving inexperienced radiologists). These preliminary results, out of the main scope of the present study, show promise and deserve dedicated further studies.

The generalizability of our results may be influenced by the technique employed and various post-processing steps, which can significantly impact CBV and PSR values. Current efforts towards standardization, as suggested by Boxerman et al [11] in 2020, recommend using either full-dose preloaded 60° flip-angle or 30° without preload sequences. However, considering the recent introduction of these recommendations and the advancements in genetic classification as per the WHO 2021 guidelines, there is currently scarce comprehensive data. This scarcity impedes the comparison of our results with others obtained using the agreed-upon consensus sequences for our specific purpose. Speculatively, using these preloaded or low FA techniques could result in higher CBV and lower PSR values, by minimizing the T1 leakage effects to optimize nrCBV measures [11], provided other post-processing steps remain consistent[12]. However, the expected differences in the studied tumors should be minor than those seen in tumors with extensive BBB disruption, and ultimately, we also applied post-processing leakage correction [47]. Additionally, non-preloaded intermediate-high flip-angle sequences (such as used in this study) have proven useful and may be the preferred choice for some authors and clinicians, specifically for pre-surgical diagnosis [13, 16, 17, 48]. Lastly, our methodology should be adaptable to different DSC-PWI pulse-sequence parameters by simply adjusting the thresholds [14]. Regardless, further multicentric validations are needed.

On another hand, our study design grouping grades 2 and 3 is a common approach in radiological papers (usually under the term “lower grade gliomas”) [36] and it is supported by the absence of any significant difference between grades in this work, which is also aligned with other radiological studies [49, 50]. Furthermore, several clinical papers also suggest that the precise genetic classification of tumors has reduced the impact of histological grading on biological behaviors [51‐54]. All these facts ensure the study objectives and results are the most clinically relevant according to current neuro-oncological diagnostic trends. Also, this design was chosen to strictly adhere to the current WHO classification molecular criteria and, lastly, to avoid excessive dataset fragmentation. At any rate, separated data for each tumor grade is provided as supplemental material for increased clarity and transparency of results.

Some alternative investigations leveraged radiomic features derived from DSC-PWI, yielding promising results with innovative methodology [55]. In contrast, our study concentrates on the specific differentiation between astrocytoma and oligodendroglioma, and employs well-established and clinically accessible metrics, which facilitates ease of understanding and practical application while forges a direct connection with the foundational histological and biological processes, potentially opening new research avenues.

This study has some limitations that must be considered. First, it is a single-site, retrospective study. However, this ensures data homogeneity, which is important for pilot and exploratory studies such as ours. Also, other DSC-PWI metrics such as mean transit time and cerebral blood flow may have a role, yet their implementation requires arterial input function determination, leading to increased complexity in calculations. In contrast, our study boasts notable strengths that deserve mention. First, all tumors are classified according to the most recent 2021 WHO Classification criteria. Our findings offer valuable clinically relevant and applicable insights from various perspectives. For instance, our study reinforces the necessity of semiautomatic and unsupervised assessment of the entire tumor percentile values, instead of assuming that mean or maximum values are the best options, a common practice in clinical settings. Notably, this observation could serve as a suggestion for software vendors to consider including these options in their packages. Additionally, we provide new insights into oligodendrogliomas’ vascular and microvascular features using a widely available clinical sequence. Moreover, our findings may be applied to other clinical questions, potentially opening new research avenues. In this regard, we acknowledge emerging DSC sequences and metrics that may offer improved information regarding permeability or vascular architecture. Nevertheless, these sequences have not yet been widely adopted in clinical practice. Ultimately, the foundational knowledge uniquely presented in this work paves the ground for their further implementation.

In conclusion, oligodendrogliomas demonstrate a higher nrCBV and lower PSR than astrocytomas, characteristics that are accentuated when voxel-wise percentile values are considered. We hypothesize that the lower PSR might reflect higher microvascular tortuosity and aligns with the well-established histological substrate of oligodendrogliomas, known as the chicken-wire capillary pattern. Lastly, the combination of nrCBV and PSR percentiles augments the value of standard single-metric and single-value evaluations, yielding promising classification results.