Per Nyman

Refined detection methods, more detailed tumor characterization, and adequate distinction between different pediatric tumor subtypes are necessary to improve diagnosis and treatment, enable precision medicine, and advance patient prognosis. However, the application of computational approaches to pediatric brain tumors remains limited, largely due to the lack of accessible datasets. To address part of this gap, we provide whole slide images (WSIs) of hematoxylin and eosin (H&E)-stained tissue sections from all pediatric central nervous system (CNS) samples collected in Sweden between 2013 and 2023. These data represent a population-based national cohort encompassing all six pediatric oncology centers in Sweden and are available through the Swedish Childhood Tumor Biobank (BTB). The dataset includes 1,446 WSIs of sufficient image quality with confirmed CNS tumor diagnoses, derived from 537 unique subjects (562 cases). In addition, diagnosticrelevant clinical information is included. Corresponding whole-genome sequencing (WGS), wholetranscriptome sequencing (WTS), and methylation array data are available for most tumor samples through separate resources. This H&E dataset has been specifically curated to support artificial intelligence-based analyses, while also serving broader applications in medical research and education. When combined with matched molecular data, it provides a valuable resource for advancing multimodal and precision diagnostic approaches in the pediatric population.

A bstract: The classification of pediatric brain tumors is investigated using deep learning on hematoxylin and eosin (H&E) and antigen Ki-67 (Ki-67) whole slide images (WSIs) from the Children’s Brain Tumor Network (CBTN) dataset. A total of 1,662 unregistered WSIs (1,047 H&E and 615 Ki-67 images) were analyzed, including low-grade glioma/astrocytoma (grades 1, 2) (LGG), high-grade glioma/astrocytoma (grades 3, 4) (HGG), medulloblastoma (MB), ependymoma (EP) and ganglioglioma. The The aim of this study was to effectively classify pediatric brain tumors using H&E and Ki-67 WSIs individually, and to investigate whether early, intermediate, and late fusion could improve the predictive performance. From each WSI, 224× 224 pixel patches were extracted, and the instance (patch)-level features were obtained using the histology foundation model CONCHv1_5. The instances were aggregated using clustering-constrained attention multiple instance learning (CLAM) for patient-level classification. Model interpretability and explainability was assessed through attention heatmaps, cell density and Ki-67 labelling index (LI) maps. In the binary grade classification between LGG and HGG, the intermediate concatenation fusion achieved the best performance with a balanced accuracy of 0.88 ± 0.05, ( p < 0.005) compared to the single-stain models (H&E: 0.84 ± 0.05, Ki-67: 0.86 ± 0.05). For the 5-class tumor type classification, the one-hidden layer late fusion learning model achieved the highest balanced accuracy of 0.83 ± 0.04 ( p < 0.005), outperforming the single-stain models (H&E: 0.77 ± 0.05, Ki-67: 0.74 ± 0.05). Overall, most of the fusion approaches outperformed the single-stain models in both classification tasks ( p < 0.005). The Ki-67 attention maps demonstrated moderate to strong Spearman correlation ( ρ = 0.576 − 0.823) with the cell density and Ki-67 LI maps, suggesting that these features are associated with the model’s predictions, although additional features may contribute. The results show that H&E and Ki-67 images provide complementary information, and most of the multi-stain fusion approaches using deep learning improve pediatric brain tumor diagnosis.

Quantification of the Kiel 67 (Ki-67) labeling index (LI) is critical for assessing proliferation and prognosis in tumors but manual scoring remains a common practice. We present an automated framework for Ki-67 scoring in whole slide images (WSIs) developed for research settings using an Apache Groovy code script for QuPath and complemented by a Python postprocessing script that provides cell density maps and summary tables. Tissue segmentation is performed by pixel classifiers and cell segmentation is conducted using StarDist, a deep learning model, followed by adaptive thresholding to classify Ki-67 positive and negative nuclei. The pipeline was applied to a cohort of 632 pediatric brain tumor cases with 734 Ki-67 WSIs from the Children’s Brain Tumor Network. Medulloblastomas showed the highest Ki-67 LI (median: 19.84), followed by atypical teratoid rhabdoid tumors (median: 19.36), brainstem glioma-diffuse intrinsic pontine gliomas (median: 11.50), high-grade gliomas (grades 3, 4) (median: 9.50), and ependymomas (median: 5.88). Lower indices were found in meningiomas (median: 1.84) and the lowest were seen in low-grade gliomas (grades 1, 2) (median: 0.85), dysembryoplastic neuroepithelial tumors (median: 0.63), and gangliogliomas (median: 0.50). The results demonstrate a significant correlation (P < .05) in Ki-67 LI across most of the tumor families/types aligning with neuro-oncology and neuropathology consensus.

Purpose: • To investigate if R ₁ and R ₂ can reliably be measured using 3D quantitative MRI in an intraoperative setting when paediatric brain tumour surgery is performed. • To determine whether B ₁ ⁺ inhomogeneities affect R ₁ and R ₂ measurements in normal-appearing white matter and the thalamus, respectively, and how R ₁ and R ₂ measurements are affected by different coils. • To assess how the relaxation parameters of brain tissue are affected by the intraoperative setting. Methods: The accuracy of R ₁ and R ₂ , the effect of B ₁ ⁺ -field inhomogeneity and how the flex coil position affected R ₁ and R ₂ were evaluated, both pre- and intraoperatively during surgery. Ten patients were recruited, six girls and four boys aged 2–15 years, with varying tumour entities, all referred to surgery with intraoperative MR. The patients were scanned using a head coil preoperatively and flex coils intraoperatively. Control experiments were performed on phantoms in various positions, equivalent to the patient positions. ROIs (Regions of Interest) were positioned in areas representing normal-appearing matter. Relaxation rates R ₁ and R ₂ were calculated from 3D-quantification using an interleaved Look-Locker acquisition sequence with T ₂ preparation pulse (3D-QALAS) data. Results: There was a significant increase of R ₂ in the intraoperative setting compared to the preoperative 3D-QALAS measurements. In contrast to the patient examinations, control experiments using relaxation phantoms did not demonstrate similar differences. Conclusion: Relaxometry is feasible in the intraoperative setting. The detected differences between the quantitative R ₂ values in tissue pre- and intraoperatively seem to be explained by the physiological conditions characterising the surgical situation.

Background: Postoperative speech impairment (POSI) and cranial nerve deficits (CND) are common complications of pediatric posterior fossa (PF) tumor surgery. Intraoperative MRI (ioMRI) has proven a useful tool in achieving gross total resection. The risk of POSI and CND with ioMRI remains unclear, making it the primary scope of this study. Additionally, we assessed whether POSI was associated with CND. Methods: We prospectively included pediatric patients undergoing PF tumor surgery in 36 centers across 15 European countries. Neurological status and speech were assessed preoperatively and 1–4 weeks postoperatively. Surgical details, including tumor location and use of ioMRI, were recorded within 72 h of surgery. Postoperative CND were categorized as 0, 1, 2, or ≥ 3 nerves affected; POSI as habitual, reduced speech, or mutism. Proportional odds models estimated odds ratios (OR) for 1) POSI with stepwise adjustment for tumor location and age, and 2) CND with adjustment for preoperative CND and tumor location. Subgroup analyses assessed systematic differences, missing data, center-level effects, and histology adjustment. Results: Of 790 primary PF tumor surgeries, 141 (18%) involved ioMRI. POSI occurred in 183/790 (23%) and postoperative CND in 213/790 (27%). POSI-risk with ioMRI showed non-significant unadjusted OR (95% CI) 0.83 (0.53;1.30); adjusted OR 0.76 (0.43;1.35). Fewer CNDs were observed with ioMRI (unadjusted OR 0.63 (0.40;1.00), adjusted OR 0.58 (0.33;0.94), p = 0.03). POSI-risk was associated with more CNDs (adjusted OR for 1 CND: 2.06 (1.15;3.68); 2 CND: 2.13 (1.02;4.42); ≥ 3 CND: 4.15 (1.98;8.70), p < 0.05). Conclusions: ioMRI was not associated with increased risk of postoperative complications in this multicenter cohort. The reduction in CND among ioMRI cases may reflect derived effects on surgical decision-making, expertise, case-load and case-mix. Results should be interpreted with caution due to limited intraoperative data. The association between POSI-risk and cumulative CND may indicate extensive brainstem involvement. Our findings highlight the need to further explore how ioMRI-guided strategies affect functional outcomes in pediatric PF tumour surgery. Clinical Trials ID: NCT02300766 (October 2014)

Purpose: To investigate if R ₁ and R ₂ can reliably be measured using 3D quantitative MRI in an intraoperative setting when paediatric brain tumour surgery is performed. To determine whether B ₁₊ inhomogeneities affect R ₁ and R ₂ measurements in normal-appearing white matter and the thalamus, respectively, and how R ₁ and R ₂ measurements are affected by different coils. To assess how the relaxation parameters of brain tissue are affected by the intraoperative setting. Methods: The accuracy of R ₁ and R ₂ , the effect of B ₁₊ -field inhomogeneity and how the flex coil position affected R ₁ and R ₂ were evaluated, both pre- and intraoperatively during surgery. Ten patients were recruited, six girls and four boys aged 2-15 years, with varying tumour entities, all referred to surgery with intraoperative MR. The patients were scanned using a head coil preoperatively and flex coils intraoperatively. Control experiments were performed on phantoms in various positions, equivalent to the patient positions. ROIs (Regions of Interest) were positioned in areas representing normal-appearing matter. Relaxation rates R ₁ and R ₂ were calculated from 3D-quantification using an interleaved Look-Locker acquisition sequence with T ₂ preparation pulse (3D-QALAS) data. Results: There was a significant increase of R ₂ in the intraoperative setting compared to the preoperative 3D-QALAS measurements. In contrast to the patient examinations, control experiments using relaxation phantoms did not demonstrate similar differences. Conclusion: Relaxometry is feasible in the intraoperative setting. The detected differences between the quantitative R ₂ tissue values pre- and intraoperatively seem to be explained by the physiological conditions characterising the surgical situation.

Purpose: Postoperative speech impairment (POSI) is a core symptom of cerebellar mutism syndrome (CMS) and is a common complication after the resection of paediatric posterior fossa (PF) tumours. Preoperative glucocorticoids (pGC) are considered standard treatment to reduce tumour oedema; in addition, glucocorticoids are often administered intraoperatively (iGC) to reduce both postoperative nausea and vomiting. The study aims to investigate whether the occurrence of POSI may be associated with pGC and iGC. Methods: In a prospective observational multicentre study, we included children with a PF tumour requiring either resection or open biopsy. The use of pGC and iGC, including drug type and dose, was registered. Postoperative speech status was classified as mutism, reduced speech, or habitual speech, where mutism and reduced speech were considered POSI of higher and lower severity, respectively. Proportional odds logistic regression with adjustment for tumour type, tumour location, and age was used to analyse the occurrence of POSI associated with glucocorticoids (GC). Results: From August 2014 to November 2024, we recruited 810 children, of whom 605 were included in the primary analysis. We found no association between the use of GC (pGC nor iGC) and the occurrence of POSI. The result did not change when adjusting for tumour type, tumour location, and age. The analysis included both a comparison between using and not using pGC (OR 1.06 [95% CI 0.46 –2.49], reference level: use of pGC) and/or iGC (1.28 [0.58–2.82], reference level: use of iGC), and a dose–response analysis of the occurrence of POSI in relation to doubling the dose of GC (pGC OR 1.28 [0.84–1.98]; iGC OR 1.07 [0.62–1.82]). Conclusion: Our study did not find evidence of a significant change in the occurrence of POSI with the use of pGC or iGC, but our results alone cannot rule out that the administration of pGC or iGC may have some effect. Therefore, our data do not call for a change in recommendations for the use of GC as protection against the development of POSI. Trial registration number: Clinicaltrials.gov (NCT02300766). Date of registration: November 25, 2014

A bstract: The quantification of the Ki-67 labeling index (LI) is critical for assessing tumor proliferation and prognosis in tumors, yet manual scoring remains a common practice. This study presents an automated workflow for Ki-67 scoring in whole slide images (WSIs) using an Apache Groovy code script for QuPath, complemented by a Python-based post-processing script, providing cell density maps and summary tables. The tissue and cell segmentation are performed using StarDist, a deep learning model, and adaptive thresholding to classify Ki-67 positive and negative nuclei. The pipeline was applied to a cohort of 632 pediatric brain tumor cases with 734 Ki-67-stained WSIs from the Children’s Brain Tumor Network. Medulloblastoma showed the highest Ki-67 LI (median: 19.84), followed by atypical teratoid rhabdoid tumor (median: 19.36). Moderate values were observed in brainstem glioma-diffuse intrinsic pontine glioma (median: 11.50), high-grade glioma (grades 3 & 4) (median: 9.50), and ependymoma (median: 5.88). Lower indices were found in meningioma (median: 1.84), while the lowest were seen in low-grade glioma (grades 1 & 2) (median: 0.85), dysembryoplastic neuroepithelial tumor (median: 0.63), and ganglioglioma (median: 0.50). The results aligned with the consensus of the oncology, demonstrating a significant correlation in Ki-67 LI across most of the tumor families/types, with high malignancy tumors showing the highest proliferation indices and lower malignancy tumors exhibiting lower Ki-67 LI. The automated approach facilitates the assessment of large amounts of Ki-67 WSIs in research settings.

Background: Cerebellar Mutism Syndrome (CMS) is a neurological complication of posterior fossa (PF) tumour surgery in children, and postoperative speech impairment (POSI) is the cardinal symptom of CMS. The role of tumour volume on the risk of POSI remains unexplored. This study investigates the association between tumour volume and the risk of POSI. Methods: We included 360 patients from the European CMS study with available preoperative T1-weighted contrast-enhanced brain MRI. Speech status was assessed within two weeks postoperatively and categorised into three levels: habitual speech, severely reduced speech, and mutism. Tumour volumes were calculated using the BrainLab Elements SmartBrush™, a semi-automated segmentation tool. We used proportional odds models to estimate the odds ratio (OR) with adjustments for tumour location, pathology, and age. Based on the primary analysis, a risk stratification model for medulloblastoma patients was constructed, and the optimal volume cut-off was determined with Youden’s Index. Results: We found no effect of the overall tumour volume on the risk of POSI. This result did not change when adjusted for tumour location, pathology, and age. We found an association between tumour volume of medulloblastoma and the risk of POSI (unadjusted OR of 1.04 per increase in cm³ (95% CI 1.01;1.07, p = 0.01)), which did not change when adjusting for tumour location and age. The risk stratification cut-off for the tumour volume of medulloblastoma was calculated to be 16,5 cm³. Patients with medulloblastoma and preoperative tumour volumes below 16,5 cm³ had an absolute risk of 13% for POSI (low-risk group), whereas patients with preoperative tumour volumes above 16,5 cm³ had an absolute risk of 50% for POSI (high-risk group). Conclusion: Our data showed an association between preoperative tumour volume and the risk of POSI in children with medulloblastoma, while no association was found for the volume of other tumour types. We suggest a straightforward cut-off risk model for assessing the risk of POSI in children with medulloblastoma based on preoperative tumour volume. This approach can aid clinicians in informing patients and parents about the complications related to CMS following PF tumour surgery in children. Clinical Trials: ID NCT02300766 (October 2014).

Purpose: To implement and evaluate deep learning-based methods for the classification of pediatric brain tumors (PBT) in magnetic resonance (MR) data. Methods: A subset of the “Children’s Brain Tumor Network” dataset was retrospectively used (n = 178 subjects, female = 72, male = 102, NA = 4, age range [0.01, 36.49] years) with tumor types being low-grade astrocytoma (n = 84), ependymoma (n = 32), and medulloblastoma (n = 62). T1w post-contrast (n = 94 subjects), T2w (n = 160 subjects), and apparent diffusion coefficient (ADC: n = 66 subjects) MR sequences were used separately. Two deep learning models were trained on transversal slices showing tumor. Joint fusion was implemented to combine image and age data, and 2 pre-training paradigms were utilized. Model explainability was investigated using gradient-weighted class-activation mapping (Grad-CAM), and the learned feature space was visualized using principal component analysis (PCA). Results: The highest tumor-type classification performance was achieved when using a vision transformer model pre-trained on ImageNet and fine-tuned on ADC images with age fusion (Matthews correlation coefficient [MCC]: 0.77 ± 0.14, Accuracy: 0.87 ± 0.08), followed by models trained on T2w (MCC: 0.58 ± 0.11, Accuracy: 0.73 ± 0.08) and T1w post-contrast (MCC: 0.41 ± 0.11, Accuracy: 0.62 ± 0.08) data. Age fusion marginally improved the model’s performance. Both model architectures performed similarly across the experiments, with no differences between the pre-training strategies. Grad-CAMs showed that the models’ attention focused on the brain region. PCA of the feature space showed greater separation of the tumor-type clusters when using contrastive pre-training. Conclusion: Classification of PBT on MR images could be accomplished using deep learning, with the top-performing model being trained on ADC data, which radiologists use for the clinical classification of these tumors.