Training and Comparison of nnU-Net and DeepMedic Methods for Autosegmentation of Pediatric Brain Tumors.

BACKGROUND AND PURPOSE: Tumor segmentation is essential in surgical and treatment planning and response assessment and monitoring in pediatric brain tumors, the leading cause of cancer-related death among children. However, manual segmentation is time-consuming and has high interoperator variability, underscoring the need for more efficient methods. After training, we compared 2 deep-learning-based 3D segmentation models, DeepMedic and nnU-Net, with pediatric-specific multi-institutional brain tumor data based on multiparametric MR images. MATERIALS AND METHODS: Multiparametric preoperative MR imaging scans of 339 pediatric patients (n = 293 internal and n = 46 external cohorts) with a variety of tumor subtypes were preprocessed and manually segmented into 4 tumor subregions, ie, enhancing tumor, nonenhancing tumor, cystic components, and peritumoral edema. After training, performances of the 2 models on internal and external test sets were evaluated with reference to ground truth manual segmentations. Additionally, concordance was assessed by comparing the volume of the subregions as a percentage of the whole tumor between model predictions and ground truth segmentations using the Pearson or Spearman correlation coefficients and the Bland-Altman method. RESULTS: The mean Dice score for nnU-Net internal test set was 0.9 (SD, 0.07) (median, 0.94) for whole tumor; 0.77 (SD, 0.29) for enhancing tumor; 0.66 (SD, 0.32) for nonenhancing tumor; 0.71 (SD, 0.33) for cystic components, and 0.71 (SD, 0.40) for peritumoral edema, respectively. For DeepMedic, the mean Dice scores were 0.82 (SD, 0.16) for whole tumor; 0.66 (SD, 0.32) for enhancing tumor; 0.48 (SD, 0.27) for nonenhancing tumor; 0.48 (SD, 0.36) for cystic components, and 0.19 (SD, 0.33) for peritumoral edema, respectively. Dice scores were significantly higher for nnU-Net (P ≤ .01). Correlation coefficients for tumor subregion percentage volumes were higher (0.98 versus 0.91 for enhancing tumor, 0.97 versus 0.75 for nonenhancing tumor, 0.98 versus 0.80 for cystic components, 0.95 versus 0.33 for peritumoral edema in the internal test set). Bland-Altman plots were better for nnU-Net compared with DeepMedic. External validation of the trained nnU-Net model on the multi-institutional Brain Tumor Segmentation Challenge in Pediatrics (BraTS-PEDs) 2023 data set revealed high generalization capability in the segmentation of whole tumor, tumor core (a combination of enhancing tumor, nonenhancing tumor, and cystic components), and enhancing tumor with mean Dice scores of 0.87 (SD, 0.13) (median, 0.91), 0.83 (SD, 0.18) (median, 0.89), and 0.48 (SD, 0.38) (median, 0.58), respectively. CONCLUSIONS: The pediatric-specific data-trained nnU-Net model is superior to DeepMedic for whole tumor and subregion segmentation of pediatric brain tumors.

A longitudinal single-cell and spatial multiomic atlas of pediatric high-grade glioma.

Pediatric high-grade glioma (pHGG) is an incurable central nervous system malignancy that is a leading cause of pediatric cancer death. While pHGG shares many similarities to adult glioma, it is increasingly recognized as a molecularly distinct, yet highly heterogeneous disease. In this study, we longitudinally profiled a molecularly diverse cohort of 16 pHGG patients before and after standard therapy through single-nucleus RNA and ATAC sequencing, whole-genome sequencing, and CODEX spatial proteomics to capture the evolution of the tumor microenvironment during progression following treatment. We found that the canonical neoplastic cell phenotypes of adult glioblastoma are insufficient to capture the range of tumor cell states in a pediatric cohort and observed differential tumor-myeloid interactions between malignant cell states. We identified key transcriptional regulators of pHGG cell states and did not observe the marked proneural to mesenchymal shift characteristic of adult glioblastoma. We showed that essential neuromodulators and the interferon response are upregulated post-therapy along with an increase in non-neoplastic oligodendrocytes. Through in vitro pharmacological perturbation, we demonstrated novel malignant cell-intrinsic targets. This multiomic atlas of longitudinal pHGG captures the key features of therapy response that support distinction from its adult counterpart and suggests therapeutic strategies which are targeted to pediatric gliomas.

AutoGVP: a dockerized workflow integrating ClinVar and InterVar germline sequence variant classification.

SUMMARY: With the increasing rates of exome and whole genome sequencing, the ability to classify large sets of germline sequencing variants using up-to-date American College of Medical Genetics-Association for Molecular Pathology (ACMG-AMP) criteria is crucial. Here, we present Automated Germline Variant Pathogenicity (AutoGVP), a tool that integrates germline variant pathogenicity annotations from ClinVar and sequence variant classifications from a modified version of InterVar (PVS1 strength adjustments, removal of PP5/BP6). This tool facilitates large-scale, clinically focused classification of germline sequence variants in a research setting. AVAILABILITY AND IMPLEMENTATION: AutoGVP is an open source dockerized workflow implemented in R and freely available on GitHub at https://github.com/diskin-lab-chop/AutoGVP.

DNA methylation landscapes in DIPG reveal methylome variability that can be modified pharmacologically.

Background: Diffuse intrinsic pontine glioma (DIPG) is a uniformly lethal brainstem tumor of childhood, driven by histone H3 K27M mutation and resultant epigenetic dysregulation. Epigenomic analyses of DIPG have shown global loss of repressive chromatin marks accompanied by DNA hypomethylation. However, studies providing a static view of the epigenome do not adequately capture the regulatory underpinnings of DIPG cellular heterogeneity and plasticity. Methods: To address this, we performed whole-genome bisulfite sequencing on a large panel of primary DIPG specimens and applied a novel framework for analysis of DNA methylation variability, permitting the derivation of comprehensive genome-wide DNA methylation potential energy landscapes that capture intrinsic epigenetic variation. Results: We show that DIPG has a markedly disordered epigenome with increasingly stochastic DNA methylation at genes regulating pluripotency and developmental identity, potentially enabling cells to sample diverse transcriptional programs and differentiation states. The DIPG epigenetic landscape was responsive to treatment with the hypomethylating agent decitabine, which produced genome-wide demethylation and reduced the stochasticity of DNA methylation at active enhancers and bivalent promoters. Decitabine treatment elicited changes in gene expression, including upregulation of immune signaling such as the interferon response, STING, and MHC class I expression, and sensitized cells to the effects of histone deacetylase inhibition. Conclusions: This study provides a resource for understanding the epigenetic instability that underlies DIPG heterogeneity. It suggests the application of epigenetic therapies to constrain the range of epigenetic states available to DIPG cells, as well as the use of decitabine in priming for immune-based therapies.

Noninvasive Molecular Subtyping of Pediatric Low-Grade Glioma with Self-Supervised Transfer Learning.

Purpose To develop and externally test a scan-to-prediction deep learning pipeline for noninvasive, MRI-based BRAF mutational status classification for pediatric low-grade glioma. Materials and Methods This retrospective study included two pediatric low-grade glioma datasets with linked genomic and diagnostic T2-weighted MRI data of patients: Dana-Farber/Boston Children's Hospital (development dataset, n = 214 [113 (52.8%) male; 104 (48.6%) BRAF wild type, 60 (28.0%) BRAF fusion, and 50 (23.4%) BRAF V600E]) and the Children's Brain Tumor Network (external testing, n = 112 [55 (49.1%) male; 35 (31.2%) BRAF wild type, 60 (53.6%) BRAF fusion, and 17 (15.2%) BRAF V600E]). A deep learning pipeline was developed to classify BRAF mutational status (BRAF wild type vs BRAF fusion vs BRAF V600E) via a two-stage process: (a) three-dimensional tumor segmentation and extraction of axial tumor images and (b) section-wise, deep learning-based classification of mutational status. Knowledge-transfer and self-supervised approaches were investigated to prevent model overfitting, with a primary end point of the area under the receiver operating characteristic curve (AUC). To enhance model interpretability, a novel metric, center of mass distance, was developed to quantify the model attention around the tumor. Results A combination of transfer learning from a pretrained medical imaging-specific network and self-supervised label cross-training (TransferX) coupled with consensus logic yielded the highest classification performance with an AUC of 0.82 (95% CI: 0.72, 0.91), 0.87 (95% CI: 0.61, 0.97), and 0.85 (95% CI: 0.66, 0.95) for BRAF wild type, BRAF fusion, and BRAF V600E, respectively, on internal testing. On external testing, the pipeline yielded an AUC of 0.72 (95% CI: 0.64, 0.86), 0.78 (95% CI: 0.61, 0.89), and 0.72 (95% CI: 0.64, 0.88) for BRAF wild type, BRAF fusion, and BRAF V600E, respectively. Conclusion Transfer learning and self-supervised cross-training improved classification performance and generalizability for noninvasive pediatric low-grade glioma mutational status prediction in a limited data scenario. Keywords: Pediatrics, MRI, CNS, Brain/Brain Stem, Oncology, Feature Detection, Diagnosis, Supervised Learning, Transfer Learning, Convolutional Neural Network (CNN) Supplemental material is available for this article. © RSNA, 2024.

Use of External Control Cohorts in Pediatric Brain Tumor Clinical Trials.

Why, when, and how to consider external control cohorts in pediatric brain tumor clinical trials.

SNOT-22 subdomain outcomes following treatment for sinonasal malignancy: A prospective, multicenter study.

BACKGROUND: Patients with sinonasal malignancy (SNM) present with significant sinonasal quality of life (QOL) impairment. Global sinonasal QOL as measured by the 22-item Sinonasal Outcomes Test (SNOT-22) has been shown to improve with treatment. This study aims to characterize SNOT-22 subdomain outcomes in SNM. METHODS: Patients diagnosed with SNM were prospectively enrolled in a multi-center patient registry. SNOT-22 scores were collected at the time of diagnosis and through the post-treatment period for up to 5 years. Multivariable regression analysis was used to identify drivers of variation in SNOT-22 subdomains. RESULTS: Note that 234 patients were reviewed, with a mean follow-up of 22 months (3 months-64 months). Rhinologic, psychological, and sleep subdomains significantly improved versus baseline (all p < 0.05). Subanalysis of 40 patients with follow-up at all timepoints showed statistically significant improvement in rhinologic, extra-nasal, psychological, and sleep subdomains, with minimal clinically important difference met between 2 and 5 years in sleep and psychological subdomains. Adjuvant chemoradiation was associated with worse outcomes in rhinologic (adjusted odds ratio (5.22 [1.69-8.66])), extra-nasal (2.21 [0.22-4.17]) and ear/facial (5.53 [2.10-8.91]) subdomains. Pterygopalatine fossa involvement was associated with worse outcomes in rhinologic (3.22 [0.54-5.93]) and ear/facial (2.97 [0.32-5.65]) subdomains. Positive margins (5.74 [2.17-9.29]) and surgical approach-combined versus endoscopic (3.41 [0.78-6.05])-were associated with worse psychological outcomes. Adjuvant radiation (2.28 [0.18-4.40]) was associated with worse sleep outcomes. CONCLUSIONS: Sinonasal QOL improvements associated with treatment of SNM are driven by rhinologic, extra-nasal, psychological, and sleep subdomains.

Predictive factors for decreased baseline quality of life in patients with sinonasal malignancies.

BACKGROUND: The impact of sinonasal malignancies (SNMs) on quality of life (QOL) at presentation is poorly understood. The Sinonasal Outcome Test (SNOT-22) and University of Washington Quality of Life (UWQOL) are validated QOL instruments with distinctive subdomains. This study aims to identify factors impacting pretreatment QOL in SNM patients to personalize multidisciplinary management and counseling. METHODS: Patients with previously untreated SNMs were prospectively enrolled (2015-2022) in a multicenter observational study. Baseline pretreatment QOL instruments (SNOT-22, UWQOL) were obtained along with demographics, comorbidities, histopathology/staging, tumor involvement, and symptoms. Multivariable regression models identified factors associated with reduced baseline QOL. RESULTS: Among 204 patients, presenting baseline QOL was significantly reduced. Multivariable regression showed worse total SNOT-22 QOL in patients with skull base erosion (p = 0.02). SNOT-rhinologic QOL was worse in women (p = 0.009), patients with epistaxis (p = 0.036), and industrial exposure (p = 0.005). SNOT extranasal QOL was worse in patients with industrial exposure (p = 0.016); worse SNOT ear/facial QOL if perineural invasion (PNI) (p = 0.027). Squamous cell carcinoma pathology (p = 0.037), palate involvement (p = 0.012), and pain (p = 0.017) were associated with worse SNOT sleep QOL scores. SNOT psychological subdomain scores were significantly worse in patients with palate lesions (p = 0.022), skull base erosion (p = 0.025), and T1 staging (p = 0.023). Low QOL was more likely in the presence of PNI on UW health (p = 0.019) and orbital erosion on UW overall (p = 0.03). UW social QOL was worse if palatal involvement (p = 0.023) or PNI (p = 0.005). CONCLUSIONS: Our findings demonstrate a negative impact on baseline QOL in patients with SNMs and suggest sex-specific and symptom-related lower QOL scores, with minimal histopathology association. Anatomical tumor involvement may be more reflective of QOL than T-staging, as orbital and skull base erosion, PNI, and palate lesions are significantly associated with reduced baseline QOL.

Everolimus for Children With Recurrent or Progressive Low-Grade Glioma: Results From the Phase II PNOC001 Trial.

PURPOSE: The PNOC001 phase II single-arm trial sought to estimate progression-free survival (PFS) associated with everolimus therapy for progressive/recurrent pediatric low-grade glioma (pLGG) on the basis of phosphatidylinositol 3-kinase (PI3K)/AKT/mammalian target of rapamycin (mTOR) pathway activation as measured by phosphorylated-ribosomal protein S6 and to identify prognostic and predictive biomarkers. PATIENTS AND METHODS: Patients, age 3-21 years, with progressive/recurrent pLGG received everolimus orally, 5 mg/m2 once daily. Frequency of driver gene alterations was compared among independent pLGG cohorts of newly diagnosed and progressive/recurrent patients. PFS at 6 months (primary end point) and median PFS (secondary end point) were estimated for association with everolimus therapy. RESULTS: Between 2012 and 2019, 65 subjects with progressive/recurrent pLGG (median age, 9.6 years; range, 3.0-19.9; 46% female) were enrolled, with a median follow-up of 57.5 months. The 6-month PFS was 67.4% (95% CI, 60.0 to 80.0) and median PFS was 11.1 months (95% CI, 7.6 to 19.8). Hypertriglyceridemia was the most common grade ≥3 adverse event. PI3K/AKT/mTOR pathway activation did not correlate with clinical outcomes (6-month PFS, active 68.4% v nonactive 63.3%; median PFS, active 11.2 months v nonactive 11.1 months; P = .80). Rare/novel KIAA1549::BRAF fusion breakpoints were most frequent in supratentorial midline pilocytic astrocytomas, in patients with progressive/recurrent disease, and correlated with poor clinical outcomes (median PFS, rare/novel KIAA1549::BRAF fusion breakpoints 6.1 months v common KIAA1549::BRAF fusion breakpoints 16.7 months; P < .05). Multivariate analysis confirmed their independent risk factor status for disease progression in PNOC001 and other, independent cohorts. Additionally, rare pathogenic germline variants in homologous recombination genes were identified in 6.8% of PNOC001 patients. CONCLUSION: Everolimus is a well-tolerated therapy for progressive/recurrent pLGGs. Rare/novel KIAA1549::BRAF fusion breakpoints may define biomarkers for progressive disease and should be assessed in future clinical trials.

AutoGVP: a dockerized workflow integrating ClinVar and InterVar germline sequence variant classification.

With the increasing rates of exome and whole genome sequencing, the ability to classify large sets of germline sequencing variants using up-to-date American College of Medical Genetics - Association for Molecular Pathology (ACMG-AMP) criteria is crucial. Here, we present Automated Germline Variant Pathogenicity (AutoGVP), a tool that integrates germline variant pathogenicity annotations from ClinVar and sequence variant classifications from a modified version of InterVar (PVS1 strength adjustments, removal of PP5/BP6). This tool facilitates large-scale, clinically-focused classification of germline sequence variants in a research setting.

A multi-institutional pediatric dataset of clinical radiology MRIs by the Children's Brain Tumor Network.

Pediatric brain and spinal cancers remain the leading cause of cancer-related death in children. Advancements in clinical decision-support in pediatric neuro-oncology utilizing the wealth of radiology imaging data collected through standard care, however, has significantly lagged other domains. Such data is ripe for use with predictive analytics such as artificial intelligence (AI) methods, which require large datasets. To address this unmet need, we provide a multi-institutional, large-scale pediatric dataset of 23,101 multi-parametric MRI exams acquired through routine care for 1,526 brain tumor patients, as part of the Children's Brain Tumor Network. This includes longitudinal MRIs across various cancer diagnoses, with associated patient-level clinical information, digital pathology slides, as well as tissue genotype and omics data. To facilitate downstream analysis, treatment-naïve images for 370 subjects were processed and released through the NCI Childhood Cancer Data Initiative via the Cancer Data Service. Through ongoing efforts to continuously build these imaging repositories, our aim is to accelerate discovery and translational AI models with real-world data, to ultimately empower precision medicine for children.

Targeted gene expression profiling predicts meningioma outcomes and radiotherapy responses.

Surgery is the mainstay of treatment for meningioma, the most common primary intracranial tumor, but improvements in meningioma risk stratification are needed and indications for postoperative radiotherapy are controversial. Here we develop a targeted gene expression biomarker that predicts meningioma outcomes and radiotherapy responses. Using a discovery cohort of 173 meningiomas, we developed a 34-gene expression risk score and performed clinical and analytical validation of this biomarker on independent meningiomas from 12 institutions across 3 continents (N = 1,856), including 103 meningiomas from a prospective clinical trial. The gene expression biomarker improved discrimination of outcomes compared with all other systems tested (N = 9) in the clinical validation cohort for local recurrence (5-year area under the curve (AUC) 0.81) and overall survival (5-year AUC 0.80). The increase in AUC compared with the standard of care, World Health Organization 2021 grade, was 0.11 for local recurrence (95% confidence interval 0.07 to 0.17, P < 0.001). The gene expression biomarker identified meningiomas benefiting from postoperative radiotherapy (hazard ratio 0.54, 95% confidence interval 0.37 to 0.78, P = 0.0001) and suggested postoperative management could be refined for 29.8% of patients. In sum, our results identify a targeted gene expression biomarker that improves discrimination of meningioma outcomes, including prediction of postoperative radiotherapy responses.

Automated temporalis muscle quantification and growth charts for children through adulthood.

Lean muscle mass (LMM) is an important aspect of human health. Temporalis muscle thickness is a promising LMM marker but has had limited utility due to its unknown normal growth trajectory and reference ranges and lack of standardized measurement. Here, we develop an automated deep learning pipeline to accurately measure temporalis muscle thickness (iTMT) from routine brain magnetic resonance imaging (MRI). We apply iTMT to 23,876 MRIs of healthy subjects, ages 4 through 35, and generate sex-specific iTMT normal growth charts with percentiles. We find that iTMT was associated with specific physiologic traits, including caloric intake, physical activity, sex hormone levels, and presence of malignancy. We validate iTMT across multiple demographic groups and in children with brain tumors and demonstrate feasibility for individualized longitudinal monitoring. The iTMT pipeline provides unprecedented insights into temporalis muscle growth during human development and enables the use of LMM tracking to inform clinical decision-making.

Radio-pathomic approaches in pediatric neuro-oncology: Opportunities and challenges.

With medical software platforms moving to cloud environments with scalable storage and computing, the translation of predictive artificial intelligence (AI) models to aid in clinical decision-making and facilitate personalized medicine for cancer patients is becoming a reality. Medical imaging, namely radiologic and histologic images, has immense analytical potential in neuro-oncology, and models utilizing integrated radiomic and pathomic data may yield a synergistic effect and provide a new modality for precision medicine. At the same time, the ability to harness multi-modal data is met with challenges in aggregating data across medical departments and institutions, as well as significant complexity in modeling the phenotypic and genotypic heterogeneity of pediatric brain tumors. In this paper, we review recent pathomic and integrated pathomic, radiomic, and genomic studies with clinical applications. We discuss current challenges limiting translational research on pediatric brain tumors and outline technical and analytical solutions. Overall, we propose that to empower the potential residing in radio-pathomics, systemic changes in cross-discipline data management and end-to-end software platforms to handle multi-modal data sets are needed, in addition to embracing modern AI-powered approaches. These changes can improve the performance of predictive models, and ultimately the ability to advance brain cancer treatments and patient outcomes through the development of such models.

Analysis and Visualization of Longitudinal Genomic and Clinical Data from the AACR Project GENIE Biopharma Collaborative in cBioPortal.

International cancer registries make real-world genomic and clinical data available, but their joint analysis remains a challenge. AACR Project GENIE, an international cancer registry collecting data from 19 cancer centers, makes data from >130,000 patients publicly available through the cBioPortal for Cancer Genomics (https://genie.cbioportal.org). For 25,000 patients, additional real-world longitudinal clinical data, including treatment and outcome data, are being collected by the AACR Project GENIE Biopharma Collaborative using the PRISSMM data curation model. Several thousand of these cases are now also available in cBioPortal. We have significantly enhanced the functionalities of cBioPortal to support the visualization and analysis of this rich clinico-genomic linked dataset, as well as datasets generated by other centers and consortia. Examples of these enhancements include (i) visualization of the longitudinal clinical and genomic data at the patient level, including timelines for diagnoses, treatments, and outcomes; (ii) the ability to select samples based on treatment status, facilitating a comparison of molecular and clinical attributes between samples before and after a specific treatment; and (iii) survival analysis estimates based on individual treatment regimens received. Together, these features provide cBioPortal users with a toolkit to interactively investigate complex clinico-genomic data to generate hypotheses and make discoveries about the impact of specific genomic variants on prognosis and therapeutic sensitivities in cancer. SIGNIFICANCE: Enhanced cBioPortal features allow clinicians and researchers to effectively investigate longitudinal clinico-genomic data from patients with cancer, which will improve exploration of data from the AACR Project GENIE Biopharma Collaborative and similar datasets.

Transcriptional immunogenomic analysis reveals distinct immunological clusters in paediatric nervous system tumours.

BACKGROUND: Cancer immunotherapies including immune checkpoint inhibitors and Chimeric Antigen Receptor (CAR) T-cell therapy have shown variable response rates in paediatric patients highlighting the need to establish robust biomarkers for patient selection. While the tumour microenvironment in adults has been widely studied to delineate determinants of immune response, the immune composition of paediatric solid tumours remains relatively uncharacterized calling for investigations to identify potential immune biomarkers. METHODS: To inform immunotherapy approaches in paediatric cancers with embryonal origin, we performed an immunogenomic analysis of RNA-seq data from 925 treatment-naïve paediatric nervous system tumours (pedNST) spanning 12 cancer types from three publicly available data sets. RESULTS: Within pedNST, we uncovered four broad immune clusters: Paediatric Inflamed (10%), Myeloid Predominant (30%), Immune Neutral (43%) and Immune Desert (17%). We validated these clusters using immunohistochemistry, methylation immune inference and segmentation analysis of tissue images. We report shared biology of these immune clusters within and across cancer types, and characterization of specific immune cell frequencies as well as T- and B-cell repertoires. We found no associations between immune infiltration levels and tumour mutational burden, although molecular cancer entities were enriched within specific immune clusters. CONCLUSIONS: Given the heterogeneity of immune infiltration within pedNST, our findings suggest personalized immunogenomic profiling is needed to guide selection of immunotherapeutic strategies.

Noninvasive molecular subtyping of pediatric low-grade glioma with self-supervised transfer learning.

Purpose: To develop and externally validate a scan-to-prediction deep-learning pipeline for noninvasive, MRI-based BRAF mutational status classification for pLGG. Materials and Methods: We conducted a retrospective study of two pLGG datasets with linked genomic and diagnostic T2-weighted MRI of patients: BCH (development dataset, n=214 [60 (28%) BRAF fusion, 50 (23%) BRAF V600E, 104 (49%) wild-type), and Child Brain Tumor Network (CBTN) (external validation, n=112 [60 (53%) BRAF-Fusion, 17 (15%) BRAF-V600E, 35 (32%) wild-type]). We developed a deep learning pipeline to classify BRAF mutational status (V600E vs. fusion vs. wildtype) via a two-stage process: 1) 3D tumor segmentation and extraction of axial tumor images, and 2) slice-wise, deep learning-based classification of mutational status. We investigated knowledge-transfer and self-supervised approaches to prevent model overfitting with a primary endpoint of the area under the receiver operating characteristic curve (AUC). To enhance model interpretability, we developed a novel metric, COMDist, that quantifies the accuracy of model attention around the tumor. Results: A combination of transfer learning from a pretrained medical imaging-specific network and self-supervised label cross-training (TransferX) coupled with consensus logic yielded the highest macro-average AUC (0.82 [95% CI: 0.70-0.90]) and accuracy (77%) on internal validation, with an AUC improvement of +17.7% and a COMDist improvement of +6.4% versus training from scratch. On external validation, the TransferX model yielded AUC (0.73 [95% CI 0.68-0.88]) and accuracy (75%). Conclusion: Transfer learning and self-supervised cross-training improved classification performance and generalizability for noninvasive pLGG mutational status prediction in a limited data scenario.

Expert-level pediatric brain tumor segmentation in a limited data scenario with stepwise transfer learning.

Purpose: Artificial intelligence (AI)-automated tumor delineation for pediatric gliomas would enable real-time volumetric evaluation to support diagnosis, treatment response assessment, and clinical decision-making. Auto-segmentation algorithms for pediatric tumors are rare, due to limited data availability, and algorithms have yet to demonstrate clinical translation. Methods: We leveraged two datasets from a national brain tumor consortium (n=184) and a pediatric cancer center (n=100) to develop, externally validate, and clinically benchmark deep learning neural networks for pediatric low-grade glioma (pLGG) segmentation using a novel in-domain, stepwise transfer learning approach. The best model [via Dice similarity coefficient (DSC)] was externally validated and subject to randomized, blinded evaluation by three expert clinicians wherein clinicians assessed clinical acceptability of expert- and AI-generated segmentations via 10-point Likert scales and Turing tests. Results: The best AI model utilized in-domain, stepwise transfer learning (median DSC: 0.877 [IQR 0.715-0.914]) versus baseline model (median DSC 0.812 [IQR 0.559-0.888]; p<0.05). On external testing (n=60), the AI model yielded accuracy comparable to inter-expert agreement (median DSC: 0.834 [IQR 0.726-0.901] vs. 0.861 [IQR 0.795-0.905], p=0.13). On clinical benchmarking (n=100 scans, 300 segmentations from 3 experts), the experts rated the AI model higher on average compared to other experts (median Likert rating: 9 [IQR 7-9]) vs. 7 [IQR 7-9], p<0.05 for each). Additionally, the AI segmentations had significantly higher (p<0.05) overall acceptability compared to experts on average (80.2% vs. 65.4%). Experts correctly predicted the origins of AI segmentations in an average of 26.0% of cases. Conclusions: Stepwise transfer learning enabled expert-level, automated pediatric brain tumor auto-segmentation and volumetric measurement with a high level of clinical acceptability. This approach may enable development and translation of AI imaging segmentation algorithms in limited data scenarios.

Radiomics for characterization of the glioma immune microenvironment.

Increasing evidence suggests that besides mutational and molecular alterations, the immune component of the tumor microenvironment also substantially impacts tumor behavior and complicates treatment response, particularly to immunotherapies. Although the standard method for characterizing tumor immune profile is through performing integrated genomic analysis on tissue biopsies, the dynamic change in the immune composition of the tumor microenvironment makes this approach not feasible, especially for brain tumors. Radiomics is a rapidly growing field that uses advanced imaging techniques and computational algorithms to extract numerous quantitative features from medical images. Recent advances in machine learning methods are facilitating biological validation of radiomic signatures and allowing them to "mine" for a variety of significant correlates, including genetic, immunologic, and histologic data. Radiomics has the potential to be used as a non-invasive approach to predict the presence and density of immune cells within the microenvironment, as well as to assess the expression of immune-related genes and pathways. This information can be essential for patient stratification, informing treatment decisions and predicting patients' response to immunotherapies. This is particularly important for tumors with difficult surgical access such as gliomas. In this review, we provide an overview of the glioma microenvironment, describe novel approaches for clustering patients based on their tumor immune profile, and discuss the latest progress on utilization of radiomics for immune profiling of glioma based on current literature.