SVSebastian R. van der Voort

Sebastian R. van der Voort

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Sebastian R. van der Voort has a total of 105 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Towards Trustworthy AI for Glioma Diagnosis: A Task-Aware Evaluation of Uncertainty Quantification

    by Mosquera Rojas, Gonzalo Esteban, van der Voort, Sebastian R., Carolin M. Pirkl, Sandeep Kaushik, Marion Smits, Stefan Klein

    Uncertainty Quantification (UQ) is a key requirement for trustworthy AI in high-stakes medical image analysis. In this work, we evaluate UQ within a multi-task Deep Learning (DL) framework for MRI-based glioma diagnosis that performs tumor segmentation and predicts Isocitrate Dehydrogenase (IDH) mutation status, 1p/19q co-deletion status and tumor grade. We use Monte Carlo Dropout (MCD) as the primary sampling-based UQ method for the detailed task-aware analysis, obtaining predictive uncertainty and its aleatoric and epistemic components. We assess uncertainty along complementary axes: (i) MC sample convergence of uncertainty estimates and their decomposition into aleatoric and epistemic components, (ii) calibration of predictive probabilities, and (iii) operational utility of uncertainty estimates, including error detection, selective prediction, and associations with tumor segmentation performance. We also compare MCD with Deep Ensembles (DE) and Monte Carlo Deep Ensembles (MCDE) to assess whether the observed operational utility of uncertainty estimates extends beyond a single UQ method. For the tumor segmentation task, we further examine how different voxel-wise uncertainty aggregation strategies influence case-level reliability assessment, thereby explicitly accounting for task-specific uncertainty representation. Additionally, we study task interactions to quantify how tumor segmentation quality and uncertainty relate to the prediction of the tumor features. Finally, we explore whether a composite trust score integrating tumor segmentation and classification uncertainty improves error detection. Across tasks, uncertainty estimates supported meaningful error detection, while calibration depended on the dropout rate, with moderate rates yielding the most reliable predictive probabilities. Decomposition provided task-dependent interpretability but did not consistently improve error detection over predictive uncertainty alone. The comparison with DE and MCDE showed that ensemble-based uncertainty estimates provided comparable operational utility, although no UQ method consistently dominated across all tasks and metrics. Moreover, the proposed trust score did not consistently outperform classification uncertainty for selective prediction, indicating that task-specific predictive uncertainty remains the most informative operational indicator of trust. Overall, our results provide a task-aware evaluation strategy and practical guidance towards the development of trustworthy AI for glioma diagnosis.

  • Open Access

    MRI-Based Assessment of Brain Tumor Hypoxia: Correlation with Histology

    by Fatemeh Arzanforoosh, Maaike Van der Velden, Avery J. L. Berman, Sebastian R. Van der Voort, Eelke M. Bos, Joost W. Schouten, Arnaud J. P. E. Vincent, Johan M. Kros, Marion Smits, Esther A. H. Warnert

    Cerebral hypoxia significantly impacts the progression of brain tumors and their resistance to radiotherapy. This study employed streamlined quantitative blood-oxygen-level-dependent (sqBOLD) MRI to assess the oxygen extraction fraction (OEF)—a measure of how much oxygen is being extracted from vessels, with higher OEF values indicating hypoxia. Simultaneously, we utilized vessel size imaging (VSI) to evaluate microvascular dimensions and blood volume. A cohort of ten patients, divided between those with glioma and those with brain metastases, underwent a 3 Tesla MRI scan. We generated OEF, cerebral blood volume (CBV), and vessel size maps, which guided 3–4 targeted biopsies per patient. Subsequent histological analyses of these biopsies used hypoxia-inducible factor 1-alpha (HIF-1α) for hypoxia and CD31 for microvasculature assessment, followed by a correlation analysis between MRI and histological data. The results showed that while the sqBOLD model was generally applicable to brain tumors, it demonstrated discrepancies in some metastatic tumors, highlighting the need for model adjustments in these cases. The OEF, CBV, and vessel size maps provided insights into the tumor’s hypoxic condition, showing intertumoral and intratumoral heterogeneity. A significant relationship between MRI-derived measurements and histological data was only evident in the vessel size measurements (r = 0.68, p < 0.001).


  • Open Access

    Microvasculature Features Derived from Hybrid EPI MRI in Non-Enhancing Adult-Type Diffuse Glioma Subtypes

    by Fatemeh Arzanforoosh, Sebastian R. van der Voort, Fatih Incekara, Arnaud Vincent, Martin Van Den Bent, Johan M. Kros, Marion Smits, Esther A. H. Warnert

    In this study, we used the vessel size imaging (VSI) MRI technique to characterize the microvasculature features of three subtypes of adult-type diffuse glioma lacking enhancement. Thirty-eight patients with confirmed non-enhancing glioma were categorized into three subtypes: Oligo (IDH-mut&1p/19q-codeleted), Astro (IDH-mut), and GBM (IDH-wt). The VSI technique provided quantitative maps of cerebral blood volume (CBV), microvasculature (µCBV), and vessel size for each patient. Additionally, tissue samples of 21 patients were histopathologically analyzed, and microvasculature features were quantified. Both MRI- and histology-derived features were compared across the three glioma subtypes with ANOVA or Kruskal–Wallis tests. Group averages of CBV, μCBV, and vessel size were significantly different between the three glioma subtypes (p < 0.01). Astro (IDH-mut) had a significantly lower CBV and µCBV compared to Oligo (IDH-mut&1p/19q-codeleted) (p = 0.004 and p = 0.001, respectively), and a higher average vessel size compared to GBM (IDH-wt) (p = 0.01). The histopathological analysis showed that GBM (IDH-wt) possessed vessels with more irregular shapes than the two other subtypes (p < 0.05). VSI provides a good insight into the microvasculature characteristics of the three adult-type glioma subtypes even when lacking enhancement. Further investigations into the specificity of VSI to differentiate glioma subtypes are thus warranted.


  • Top Cited Papers
    Open Access

    Combined molecular subtyping, grading, and segmentation of glioma using multi-task deep learning

    by Sebastian R van der Voort, Fatih Incekara, Maarten M J Wijnenga, Georgios Kapsas, Renske Gahrmann, Joost W Schouten, Rishi Nandoe Tewarie, Geert J Lycklama, Philip C De Witt Hamer, Roelant S Eijgelaar, et al.

    Background: Accurate characterization of glioma is crucial for clinical decision making. A delineation of the tumor is also desirable in the initial decision stages but is time-consuming. Previously, deep learning methods have been developed that can either non-invasively predict the genetic or histological features of glioma, or that can automatically delineate the tumor, but not both tasks at the same time. Here, we present our method that can predict the molecular subtype and grade, while simultaneously providing a delineation of the tumor. Methods: We developed a single multi-task convolutional neural network that uses the full 3D, structural, preoperative MRI scans to predict the IDH mutation status, the 1p/19q co-deletion status, and the grade of a tumor, while simultaneously segmenting the tumor. We trained our method using a patient cohort containing 1508 glioma patients from 16 institutes. We tested our method on an independent dataset of 240 patients from 13 different institutes. Results: In the independent test set, we achieved an IDH-AUC of 0.90, an 1p/19q co-deletion AUC of 0.85, and a grade AUC of 0.81 (grade II/III/IV). For the tumor delineation, we achieved a mean whole tumor Dice score of 0.84. Conclusions: We developed a method that non-invasively predicts multiple, clinically relevant features of glioma. Evaluation in an independent dataset shows that the method achieves a high performance and that it generalizes well to the broader clinical population. This first-of-its-kind method opens the door to more generalizable, instead of hyper-specialized, AI methods.


  • Open Access

    DeepDicomSort: An Automatic Sorting Algorithm for Brain Magnetic Resonance Imaging Data

    by Sebastian R. van der Voort, Marion Smits, Stefan Klein, For The Alzheimer’s Disease Neuroimaging Initiative

    With the increasing size of datasets used in medical imaging research, the need for automated data curation is arising. One important data curation task is the structured organization of a dataset for preserving integrity and ensuring reusability. Therefore, we investigated whether this data organization step can be automated. To this end, we designed a convolutional neural network (CNN) that automatically recognizes eight different brain magnetic resonance imaging (MRI) scan types based on visual appearance. Thus, our method is unaffected by inconsistent or missing scan metadata. It can recognize pre-contrast T1-weighted (T1w),post-contrast T1-weighted (T1wC), T2-weighted (T2w), proton density-weighted (PDw) and derived maps (e.g. apparent diffusion coefficient and cerebral blood flow). In a first experiment,we used scans of subjects with brain tumors: 11065 scans of 719 subjects for training, and 2369 scans of 192 subjects for testing. The CNN achieved an overall accuracy of 98.7%. In a second experiment, we trained the CNN on all 13434 scans from the first experiment and tested it on 7227 scans of 1318 Alzheimer’s subjects. Here, the CNN achieved an overall accuracy of 98.5%. In conclusion, our method can accurately predict scan type, and can quickly and automatically sort a brain MRI dataset virtually without the need for manual verification. In this way, our method can assist with properly organizing a dataset, which maximizes the shareability and integrity of the data.


  • Open Access

    Topographical Mapping of 436 Newly Diagnosed IDH Wildtype Glioblastoma With vs. Without MGMT Promoter Methylation

    by Fatih Incekara, Sebastian R. van der Voort, Hendrikus J. Dubbink, Peggy N. Atmodimedjo, Rishi Nandoe Tewarie, Geert Lycklama, Arnaud J. P. E. Vincent, Johan M. Kros, Stefan Klein, Martin van Den Bent, et al.

    Introduction: O6-methylguanine-methyltransferase (MGMT) promoter methylation and isocitrate dehydrogenase (IDH) mutation status are important prognostic factors for patients with glioblastoma. There are conflicting reports about a differential topographical distribution of glioblastoma with vs. without MGMT promoter methylation, possibly caused by molecular heterogeneity in glioblastoma populations. We initiated this study to re-evaluate the topographical distribution of glioblastoma with vs. without MGMT promoter methylation in light of the updated WHO 2016 classification. Methods: Preoperative T2-weighted/FLAIR and postcontrast T1-weighted MRI scans of patients aged 18 year or older with IDH wildtype glioblastoma were collected. Tumors were semi-automatically segmented, and the topographical distribution between glioblastoma with vs. without MGMT promoter methylation was visualized using frequency heatmaps. Then, voxel-wise differences were analyzed using permutation testing with Threshold Free Cluster Enhancement. Results: Four hundred thirty-six IDH wildtype glioblastoma patients were included; 211 with and 225 without MGMT promoter methylation. Visual examination suggested that when compared with MGMT unmethylated glioblastoma, MGMT methylated glioblastoma were more frequently located near bifrontal and left occipital periventricular area and less frequently near the right occipital periventricular area. Statistical analyses, however, showed no significant difference in topographical distribution between MGMT methylated vs. MGMT unmethylated glioblastoma. Conclusions: This study re-evaluated the topographical distribution of MGMT promoter methylation in 436 newly diagnosed IDH wildtype glioblastoma, which is the largest homogenous IDH wildtype glioblastoma population to date. There was no statistically significant difference in anatomical localization between MGMT methylated vs. unmethylated IDH wildtype glioblastoma.


  • Open Access

    Predicting the 1p/19q Codeletion Status of Presumed Low-Grade Glioma with an Externally Validated Machine Learning Algorithm

    by Sebastian R. van der Voort, Fatih Incekara, Maarten M.J. Wijnenga, Georgios Kapsas, Mayke Gardeniers, Joost W. Schouten, Martijn P.A. Starmans, Rishie Nandoe Tewarie, Geert J. Lycklama, Pim J. French, et al.

    Purpose: Patients with 1p/19q codeleted low-grade glioma (LGG) have longer overall survival and better treatment response than patients with 1p/19q intact tumors. Therefore, it is relevant to know the 1p/19q status. To investigate whether the 1p/19q status can be assessed prior to tumor resection, we developed a machine learning algorithm to predict the 1p/19q status of presumed LGG based on preoperative MRI. Experimental Design: Preoperative brain MR images from 284 patients who had undergone biopsy or resection of presumed LGG were used to train a support vector machine algorithm. The algorithm was trained on the basis of features extracted from post-contrast T1-weighted and T2-weighted MR images and on patients' age and sex. The performance of the algorithm compared with tissue diagnosis was assessed on an external validation dataset of MR images from 129 patients with LGG from The Cancer Imaging Archive (TCIA). Four clinical experts also predicted the 1p/19q status of the TCIA MR images. Results: The algorithm achieved an AUC of 0.72 in the external validation dataset. The algorithm had a higher predictive performance than the average of the neurosurgeons (AUC 0.52) but lower than that of the neuroradiologists (AUC of 0.81). There was a wide variability between clinical experts (AUC 0.45–0.83). Conclusions: Our results suggest that our algorithm can noninvasively predict the 1p/19q status of presumed LGG with a performance that on average outperformed the oncological neurosurgeons. Evaluation on an independent dataset indicates that our algorithm is robust and generalizable.


  • Open Access

    Radiomics approach to distinguish between well differentiated liposarcomas and lipomas on MRI

    by M Vos, M P A Starmans, M J M Timbergen, S R van der Voort, G A Padmos, W Kessels, W J Niessen, G J L H van Leenders, D J Grünhagen, S Sleijfer, et al.

    Background: Well differentiated liposarcoma (WDLPS) can be difficult to distinguish from lipoma. Currently, this distinction is made by testing for MDM2 amplification, which requires a biopsy. The aim of this study was to develop a noninvasive method to predict MDM2 amplification status using radiomics features derived from MRI. Methods: Patients with an MDM2-negative lipoma or MDM2-positive WDLPS and a pretreatment T1-weighted MRI scan who were referred to Erasmus MC between 2009 and 2018 were included. When available, other MRI sequences were included in the radiomics analysis. Features describing intensity, shape and texture were extracted from the tumour region. Classification was performed using various machine learning approaches. Evaluation was performed through a 100 times random-split cross-validation. The performance of the models was compared with the performance of three expert radiologists. Results: The data set included 116 tumours (58 patients with lipoma, 58 with WDLPS) and originated from 41 different MRI scanners, resulting in wide heterogeneity in imaging hardware and acquisition protocols. The radiomics model based on T1 imaging features alone resulted in a mean area under the curve (AUC) of 0·83, sensitivity of 0·68 and specificity of 0·84. Adding the T2-weighted imaging features in an explorative analysis improved the model to a mean AUC of 0·89, sensitivity of 0·74 and specificity of 0·88. The three radiologists scored an AUC of 0·74 and 0·72 and 0·61 respectively; a sensitivity of 0·74, 0·91 and 0·64; and a specificity of 0·55, 0·36 and 0·59. Conclusion: Radiomics is a promising, non-invasive method for differentiating between WDLPS and lipoma, outperforming the scores of the radiologists. Further optimization and validation is needed before introduction into clinical practice.


  • NIMG-75. WHO 2016 GRADE II GLIOMA MOLECULAR SUBTYPES HAVE A DISTINCT SPATIAL DISTRIBUTION PATTERN

    by Maarten Wijnenga, Sebastian van der Voort, Pim French, Stefan Klein, Hendrikus Dubbink, Winand Dinjens, Peggy Atmodimedjo, Joost Schouten, Johan Kros, Clemens Dirven, et al.

    Several studies reported a correlation between the anatomic location of gliomas and the genetic background of the tumor. As such, tumor location may contribute to pre-surgical clinical decision-making. Current evidence is mainly derived from smaller series and report single genes or copy numbers, but not the integrated WHO 2016 classification. Our purpose was to assess the location distribution of different WHO 2016 glioma subgroups in a consecutive series of molecularly defined grade II glioma and to create location heatmaps of the low grade glioma molecular subtypes. 205 adult patients with a grade II supratentorial glioma diagnosed between 2003–2016 were included in this study (100 IDH mutated astrocytoma, 85 IDH mutated and 1p19q co-deleted oligodendroglioma, and 20 IDH wildtype astrocytoma). All patients were classified according to the WHO 2016 criteria using a dedicated Next-Generation-Sequencing panel. Tumor volume and location were assessed with semi-automatic software on T2-weighted images. All volumes of interest were mapped to a standard brain using affine registration. Location heatmaps were created for each WHO 2016 glioma subgroup by overlaying segmentations on a standard brain. Our location maps confirmed earlier observations that different molecular subtypes of grade II glioma have a different spatial distribution: most IDH mutated 1p19q co-deleted oligodendrogliomas were located in the frontal lobes and cortex, while IDH mutated astrocytomas were more frequently located in the frontotemporal and insular region. IDH wildtype astrocytomas were predominantly located in the basal ganglia and temporal lobes. Our observations were significant as confirmed with voxel-based Fisher’s exact tests (figures will be shown at SNO). Each WHO 2016 glioma subgroup displayed a distinct pattern of spatial distribution. Our observations may contribute to pre-surgical clinical decision-making and to radiogenomics.

  • OS6.2 Identifying molecular subtypes of non-enhancing glioma using MRI perfusion and diffusion parameters

    by S Koene, F Incekara, S van der Voort, A Vincent, M van Den Bent, G Lycklama À Nijeholt, R Nandoe Tewari, M Smits

    According to the 2016 WHO classification low grade glioma can be classified into three groups based on molecular profile. A non-invasive method to identify these molecular profiles has yet to be established. There is evidence MRI perfusion and diffusion imaging might aid in the differentiation of the molecular profiles. Our aim is to identify MR imaging parameters to distinguish molecular profiles of non-enhancing glioma. 96 patients with non-enhancing, presumed low grade glioma were included. Patients were included if a pre-operative MRI-scan was available including at least T2w/FLAIR, dynamic susceptibility contrast perfusion and diffusion imaging and if the molecular status - 1p/19q co-deletion status and IDH mutation - was known. Tumors were segmented semi-automatically. Imaging parameters such as relative Cerebral Blood Volume (rCBV) and Apparent Diffusion Coefficient (ADC) were derived using IB Neuro. Tumors were divided into three clinically relevant molecular subgroups; 1p/19q co-deletion with IDH mutation (n=39), IDH-mutation without co-deletion (n=37), and IDH-wildtype without co-deletion (n=20). Kruskall-Wallace test showed significant differences for rCBV (Median, 75th and 85th percentile) and ADC (Median, 15th and 25th percentile) values between the three groups (p<0.05). Post-hoc analysis showed that IDH-mutated tumors without co-deletion had a significantly lower rCBV compared to the 1p/19q-codeleted tumors and the IDH-wildtype tumors (Median rCBV 0.67, 0.98 and 1.02 respectively p<0.05). Also, IDH-mutated, non-codeleted tumors have a significantly higher ADC compared to the two other groups (Median ADC 1.52 versus 1.27 and 1.17 p<0.05). No significant differences were found between 1p/19q co-deleted tumors and IDH-wildtype tumors in ADC or rCBV. Multinomial regression analysis (with age, tumor location and median rCBV/ADC ratio) showed an accurate prediction of the molecular profile of 79.2%. IDH-mutated, non-codeleted tumors had a significantly lower rCBV and significantly higher ADC compared to IDH-wildtype and 1p/19q co-deleted tumors. These results suggest that MRI perfusion and diffusion parameters are promising to determine mutation status of non-contrast enhancing low grade glioma.