Research Presentation Session: Neuro

RPS 2211 - Brain tumour imaging: from radiology to pathology

March 3, 08:00 - 09:00 CET

7 min
Default mode network functional connectivity is sensitive to glioma WHO-grade
Ahmed M. Radwan, Leuven / Belgium
Author Block: A. M. Radwan, H. Vandermeulen, F. Samardzic, J. M. Sousa, S. Sunaert; Leuven/BE
Purpose: This study examined the impact of glioma characteristics (WHO grade and volume), patient age, and gender on the functional connectivity (FC) of seven canonical resting-state networks (RSNs). Specifically, FC difference in the default mode (DMN), dorsal attention (DAN), fronto-parietal (FP), language (Lang), salience (SAL), sensory-motor (SMN), and visual (Vis) networks.
Methods or Background: BOLD resting-state functional magnetic resonance imaging (rs-fMRI) data were collected at 3 Tesla from 40 neurosurgery-naïve patients, newly diagnosed with cerebral glioma. Data were preprocessed using fmriprep, and denoising and seed-to-voxel functional connectivity mapping (SBA-FC) were performed with CONN. Hypothesis testing was conducted using FSL’s Randomise for nonparametric two-group t-tests. This excluded subject-specific lesioned voxels and controlled for lesion volume, patient age, and gender. SBA-FC beta maps were compared between high-grade gliomas (HGG, WHO grades III-IV, N=21) and low-grade gliomas (LGG, WHO grades I-II, N=19). Separate statistical contrasts also evaluated the effects of lesion volume, patient age, and gender.
Results or Findings: Global DMN connectivity was significantly reduced in HGG patients compared to LGG (579 voxels, t-value mean=3, PFDR<.05), involving the posterior cingulate and precuneus gyri. A small focus (18 voxels) of significantly increasing FC with age was identified in the right supplementary motor area (SMA) (t-value mean=4.45, PFDR<0.05). Tumour volume and patient gender did not significantly influence FC in any of the seven RSNs.
Conclusion: DMN functional connectivity is notably sensitive to glioma WHO grade. FC in the right SMA appeared to increase with patient age, while lesion volume and patient gender did not impact FC in the other RSNs.
Limitations: This study had a limited sample size and a lack of neuropsychological data.
Funding for this study: No funding was provided for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the UZ/KU Leuven ethics committee study number is S61759.
7 min
IDH status prediction in gliomas using machine-learning analysis of multiparametric MRI
Vojtěch Sedlák, Humpolec / Czechia
Author Block: V. Sedlák, T. Belsan, D. Netuka, A. Kavková; Prague/CZ
Purpose: This study aimed to explore the efficacy of machine-learning algorithms in accurately predicting Isocitrate Dehydrogenase (IDH) mutation status in adult-type diffuse brain gliomas, utilising quantitative data extracted from multiparametric MRI, to enhance diagnostic precision and potentially guide personalized treatment strategies.
Methods or Background: A cohort of 100 patients underwent comprehensive multimodal MRI, encompassing ASL perfusion, DSC perfusion, advanced diffusion imaging (including DKI, SMT and other models) and standard morphological imaging (i.e. T2, FLAIR, SWI, pre and postcontrast T1). Quantitative features were then extracted from these scans and fed into machine-learning algorithms, with the objective of developing a predictive model for IDH status in gliomas. Investigated algorithms included random forest, XGBoost, AdaBoost, logistic regression and support vector machine models.
Results or Findings: Various performance metrics were assessed for each model with emphasis on accuracy and AUC. The investigated machine-learning models achieved high diagnostic accuracies in determining the IDH mutation status, with areas-under-the-curve ranging from 89% for Random Forrest to 97% in the case of the Logistic Regression model.
Conclusion: The integration of machine-learning algorithms with multiparametric MRI data demonstrates a promising avenue for the accurate prediction of IDH status in glioma patients. This approach not only substantiates the pivotal role of advanced imaging techniques in diagnostic neuro-oncology but also underscores the transformative impact of machine-learning in medical diagnostics and patient stratification.
Limitations: The main limitation of the study is the still relatively modest sample size in combination with the inherent heterogeneity of glioma characteristics, which in combination might introduce potential bias in algorithm training. Further studies with larger cohorts and external validation are imperative to ascertain the generalisability of these models.
Funding for this study: This research was financially supported by the Charles University Grant Agency (project no. 222623) entitled “Advanced Diffusion MR Imaging in Diagnosis of Brain Tumors”, implemented at the Second Faculty of Medicine of Charles University.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Ethics Committee of the Military University Hospital in Prague.
7 min
DSC-PWI presurgical differentiation of Grade 4 astrocytoma and glioblastoma in young adults: unsupervised percentilic rCBV analysis across enhancing and non-enhancing regions
Ady Mildred Viveros, Barcelona / Spain
Author Block: A. M. Viveros, P. Naval-Baudin, S. Flores Casaperalta, F. A. Garay Buitron, S. Septién Rivera, M. Cos Domingo, C. Majós, A. Pons Escoda; Barcelona/ES
Purpose: This research aimed to evaluate the differentiation ability of relative-cerebral-blood-volume (rCBV) percentile values for the enhancing and non-enhancing tumour regions compared to the more commonly used mean or maximum preselected rCBV values. The presurgical differentiation between IDH-mutant astrocytoma-grade-4 and IDH-wildtype-glioblastoma is relevant for patient management, especially in young adults. It provides prognostic information and aids in guiding the molecular diagnostic work-up or in identifying patients for trials on IDH-directed treatments. While DSC-PWI has demonstrated potential for this task, its full capabilities may not yet have been realised.
Methods or Background: Patients with grade 4 astrocytic tumours, known IDH-mutation status, available presurgical MR with DSC-PWI, and under 55 years old (threshold below which IDH-mutations are evenly balanced) were retrospectively retrieved from 2016- 2023. Both enhancing and non-enhancing regions were 3D-segmented. Voxel-level rCBV was calculated to derive mean, maximum, and percentile values. Statistical comparisons were performed using the Mann-Whitney U test and AUC-ROC.
Results or Findings: The study comprised 59 patients: 11 astrocytoma-4, and 48 glioblastoma. The enhancing regions of glioblastoma displayed a higher rCBV, though the differences were not statistically significant. The non-enhancing components of astrocytoma-4 exhibited significantly higher rCBV, more pronounced when assessing lower percentiles. The 30th rCBV percentile values for the non-enhancing region were 0.705 in astrocytoma-4 and 0.458 in glioblastoma, with a p-value of 0.001 and AUC-ROC of 0.811. This outperformed the results derived from the commonly used mean and maximum.
Conclusion: An unsupervised percentile-based approach to select rCBV values enhances the differentiation outcomes over the traditional mean and maximum. The non-enhancing region offers more valuable insights than the enhancing region. Elevated rCBV values in the lower percentiles of the non-enhancing component in astrocytoma-4 are the most distinguishable characteristic and may represent very-low vascularized infiltrated tissue, versus pure oedema in glioblastoma.
Limitations: This was a single-site and retrospective investigation.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Ethics Committee of Hospital Universitari de Bellvitge.
7 min
Tumour-ipsilateral hemisphere T2 relaxometry predicts progression-free survival in patients with primary glioblastoma
Josef Vymazal, Prague / Czechia
Author Block: J. Vymazal, A. Rulseh; Prague/CZ
Purpose: This study aimed to assess whether repeated MR T2 relaxometry of the tumour-ipsilateral hemisphere can predict the progression-free survival (PFS) of patients with primary glioblastoma (GBM).
Methods or Background: 299 MR examinations in 32 GBM patients were included in this study. T2 in the tumour ipsilateral hemisphere was calculated and plotted against PFS. The evaluation period ranged between 4 months and 19 years. Patients with no disease progression were included only if they exceeded 2.5 years of follow-up; only data from the first 2.5 years were used if PFS was longer. 1/T2 was plotted against PFS with linear regression and the slope was calculated. No post-progression data were used.
Results or Findings: Seven patients with PFS longer than 6 years (6-19 years, three of them still with no progression) had an average slope of 0.0018 (-0.1294 to 0.4256). Seven patients with PFS less than 1 year (0.329 to 0.944) had an average slope of -1.2659 (-3.7109 to -0.1964), p=0.025. Eight patients with PFS between 1-2 years had an average slope of -0.619 (-0.1412 to -1,3223) p=0. Ten patients with PFS between 2-5 years had an average slope of -0.4032 (-0.6595 to -0.1331), compared with PFS longer than 6 years p=0.0015.
Conclusion: T2 relaxometry from the tumour-ipsilateral hemisphere reliably predicted PFS longer than 6 years, based on data from the first 2.5 years. The average linear regression slope was dependent on PFS: No patients with PFS longer than 6 years had a slope lower than -0.12, and no patients with a slope less than -0.8 had PFS longer than 1.67 years. This methodology may select patients with a high risk of early recurrence and also those where long-term PFS can be expected.
Limitations: This was a partially retrospective study.
Funding for this study: This study was supported by MH CZ – DRO (Na Homolce Hospital – NNH, 00023884) IG204301, IG204302.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Ethics Committee of the Na Homolce Hospital, Prague, The Czech Republic.
7 min
Quantitative histopathological analysis of the periphery of glioblastoma based on ADC and rCBV
Juan Romero Coronado, Madrid / Spain
Author Block: J. Romero Coronado, A. Ramos Gonzalez, A. Hilario Barrio, E. Salvador Alvarez, C. Lechuga Vázquez, Z. H. Chen Zhou, A. C. Martinez De Aragón Calvo, A. Cardenas; Madrid/ES
Purpose: In glioblastoma, non-enhancing areas with FLAIR hyperintensity represent both vasogenic oedema and tumour infiltration. The purpose of this study is to correlate the degree of tumour infiltration in histological samples obtained from the FLAIR hyperintense area, targeted by ADC and rCBV in a pre-surgical analysis.
Methods or Background: A total of 33 biopsies performed on 11 patients diagnosed with glioblastoma were analysed. Samples were obtained in their first surgical procedure in patients without any prior treatment.
MRI with DWI and DSC sequences were performed and analyzed with Olea Sphere software. We obtained different quantitative parameters using ROIs in the periphery of the non-enhancing tumour. During surgery, before the enhancing tumour was removed, biopsies were obtained from the periphery, previously targeted based on ADC and rCBV. To assess the proliferative potential of our samples, we performed immunohistochemical staining of p53 and MIB1. All slides underwent digital scanning using a NanoZoomer-SQ scanner, and positive cells were quantified using QuPath-0.4.2 software.
Results or Findings: In this study, we have obtained a valuable correlation between ADC and rCBV data, and the presence of tumour infiltration in the non-enhancing peripheral tumour with high signal in T2 and FLAIR sequences. The rCBV does not correlate when the value is above 1, in which case we always found tumour infiltration. The ADC correlates with infiltration when the values are low, while we have unexpectedly found areas with high ADC with dense tumour infiltration presumably due to the presence of large vasogenic oedema.
Conclusion: Our findings provide valuable insights into the nature of the peripheral zone of glioblastomas. Future studies will help us to correlate radiomic parameters with the degree of tumour infiltration, considering pathological, biological and immune biomarkers.
Limitations: No limitations were identified.
Funding for this study: This study was funded by the Fondo de Investigación en Salud (FIS).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: No information was provided by the submitter.
7 min
Differentiating brain metastases, glioblastoma and primary central nervous system lymphoma non-invasively using artificial intelligence-based multiparametric MRI
Junjie Li, Beijing / China
Author Block: J. Li, L. Chai, Z. Zhuo, Y. Duan, Y. Liu; Beijing/CN
Purpose: This study aimed to differentiate brain metastases, glioblastoma and primary central nervous system lymphoma non-invasively using artificial intelligence-based multiparametric MRI. Accurate differentiation of brain metastases (BM), glioblastoma (GBM), and primary central nervous system lymphoma (PCNSL) is crucial in clinical practice. However, most studies on artificial intelligence (AI) only focus on the differentiation of two types of tumours, lacking research on AI methods for the simultaneous differentiation of all three tumours.
Methods or Background: This study included preoperative multiparametric MRI images of BM (n=375), GBM (n=391), and PCNSL (n=361). The MRI sequences consisted of T1w, T2w, FLAIR, ADC, and contrast-enhanced T1 images. The data were randomly divided into a training set (n=788) and a test set (n=339) in a 7:3 ratio. A fully automated differentiation model was developed based on the multiparametric MRI for tumour differentiation. The results of the model were compared with those of junior and senior neuro-radiologists.
Results or Findings: The accuracy of the model in differentiating BM, GBM, and PCNSL was 0.73, with corresponding AUC values of 0.82, 0.88, and 0.88. The results were similar to those of senior neuro-radiologists (accuracy: 0.74; AUC: 0.87, 0.90, 0.90) and higher than those of junior neuro-radiologists (accuracy: 0.60; AUC: 0.60, 0.73, 0.71). The accuracy improved for the junior neuro-radiologists after re-evaluating the cases using the model's results (accuracy: 0.70; AUC: 0.71, 0.81, 0.77).
Conclusion: The model utilises multiparametric MRI for non-invasive differentiation of BM, GBM, and PCNSL. The results of the model are similar to those of senior neuro-radiologists and superior to those of junior neuro-radiologists. Thus, the diagnostic proficiency of junior neuro-radiologists was improved.
Limitations: Firstly, all data were obtained from a single centre. Additionally, The number of PCNSLs was relatively lower in all data sets, and this class imbalance may impart statistical bias in model performance. Lastly, the model was trained only with axial MRI slices.
Funding for this study: This work was supported by the Beijing Municipal Natural Science Foundation for Distinguished Young Scholars (No. JQ20035), Capital Health Development Research Project (NO. 2022-1-2042); Radiographic Standard Database Construction Project (NO. YXFSC2022JJSJ004).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Institutional Review Board of our hosipital, and written informed consent was obtained from all patients or their legal guardians.

This session will not be streamed, nor will it be available on-demand!