Kristen W. Yeom

Purpose Non-midline, pediatric high-grade gliomas (HGG) represent one of the leading causes of cancer-related deaths in children with underlying molecular genetics and driver mutations that are distinct from adult HGGs. Image-based biomarkers that assist pediatric HGG risk-stratification could potentiate treatment strategies. Towards this, we investigated radiomics approaches for non-midline pediatric HGG prognosis. Methods We identified 77 children (mean age: 140 months; 43 males) with non-midline-origin, hemispheric pediatric HGG tumors across five pediatric institutions. We extracted 1800 image-biomarker standardization initiative (IBSI)-based radiomics tumor features from treatment-naïve, axial gadolinium-enhanced axial T1- and axial T2-weighted brain MRI (Gad T1-MRI, T2-MRI). We performed k-fold cross validation and Cox regression to identify optimal predictive features of overall survival. We calculated the risk scores for each patient using a linear combination of the selected features weighted by their coefficients determined by the Cox regression model. Patients were stratified based on the median risk score into high- and low-risk groups. Python (version 3.10.0) was used for all model development. Results The Cox regression model that included both clinical (age and sex) and MRI-derived radiomics features demonstrated a concordance of 0.78 (95% CI: 0.70–0.83) compared to concordance of 0.75 (95% CI: 0.69–0.82) that used radiomics features alone and concordance of 0.61 (95% CI: 0.54–0.67) that used clinical features (age and sex) alone. Using the risk scores derived from our radiomics model, we present a Kaplan-Meier curve on our HGG cohort. Median overall survival was 21.7 months in the high-risk group and 44.6 months in the low-risk group (log-rank P = 0.007; hazard ratio 2.42, 95% CI 1.26–4.66). Conclusion In this multi-center, pilot study, we identified optimal radiomics features in the creation of a prognostic pediatric HGG model. Computational MRI techniques may offer new approaches for quantitatively evaluating tumor phenotype and serve a potential future role in therapy planning and clinical trials eligibility.

Pediatric low-grade gliomas (pLGGs) are the most common brain tumors in children and frequently harbor BRAF alterations, most commonly KIAA1549-BRAF fusions and BRAF V600E mutations, which have distinct therapeutic implications. The aim of our study was to assess multiclass radiomics-based prediction of BRAF mutation status in children with pLGG using multi-sequence MRI. This study follows TRIPOD-AI guidelines. This retrospective bi-institutional study included pediatric patients with pLGG and known BRAF mutation status who underwent pre-surgical MRI between January 2009 and January 2023. Tumors were manually segmented, and radiomics features were extracted using PyRadiomics. Random Forest classifiers were trained for three-class classification (BRAF fusion vs BRAF V600E vs non-BRAF) using clinical-only, radiomics-only, and combined models. Performance was evaluated with leave-one-out cross-validation, and results were compared across single-sequence and multisequence approaches. Single-sequence models were trained using all available patients for each MRI sequence, whereas multisequence models were restricted to the subset of 180 patients with all four sequences available. 511 children were included (mean age 8.5 ± 5.1 years; 45% female). Molecular subtypes included BRAF fusion (223/511, 44.6%), BRAF V600E (105/511, 21.0%), and non-BRAF tumors (172/511, 34.4%). FLAIR sequences were available for 495, T2WI for 454, contrast-enhanced T1WI (CE-T1WI) for 285 and ADC maps for 252 children. All sequences were available for 180 children. FLAIR was the best-performing single sequence (AUC 0.82), followed by T2WI (0.80), ADC (0.77), and CE-T1WI (0.75). Reported AUC values represent macro-average one-vs-rest performance across the three molecular classes. Combined clinical-radiomics models consistently outperformed single-source models. In the 180-patient multisequence cohort, radiomics feature concatenation (macro-AUC 0.79) and ensemble modeling (0.79) both outperformed single-sequence approaches (p < 0.001). Feature analysis showed FLAIR-derived features dominated, but adding T2, ADC, and CE-T1WI improved balanced classification across subtypes. MRI-based machine learning models may support noninvasive prediction of BRAF mutation status in pLGG. FLAIR is the best-predicting single sequence, but multisequence integration was associated with improved and more balanced performance. These findings support multisequence radiomics as a promising tool to guide precision treatment in pLGG, particularly when tissue sampling is not feasible.

Background Arterial spin labeling (ASL) MRI is a non-invasive perfusion imaging technique with potential for assessing hemodynamics in children. However, understanding hemodynamic changes in developing brains remains challenging. This study investigates the impact of normal brain development on ASL-derived cerebral blood flow (CBF) and arterial transit time (ATT) in MR-negative children. Methods Thirty-two pediatric subjects (ages 0.7–17.9 years, mean 9.3 ± 5.3y, 19 male) with MR-negative findings were retrospectively included. Four pseudo-continuous ASL (PCASL) scans were acquired: single-delay (PLD 1525 or 2025 ms) and multi-delay (3 or 7 delays). CBF and ATT in supratentorial gray matter (GM), white matter (WM), and total-brain (TB) regions were analyzed using paired t-tests, Cohen’s d, Spearman correlation, and mixed linear regression models. Results Single-delay CBF was significantly higher than 3-delay CBF in GM and WM (p < 0.001 PLD 2025 ms; p = 0.02 PLD 1525 ms). WM and TB CBF correlated negatively with age (rho=-0.56, p < 0.001), whereas GM CBF showed no significant correlation (rho=-0.03, p = 0.87); the trends differed significantly (p = 0.01). GM and TB ATT increased with age (r² > 0.11, p < 0.021). WM and TB CBF correlated with WM and combined WM/GM volumes (rho=-0.42, p = 0.02; rho=-0.46, p = 0.008). Conclusion GM and WM exhibit distinct age-related hemodynamic patterns. WM perfusion declines with age and correlates with WM volume, while GM perfusion remains stable. The progressive increase in GM ATT highlights the need for cautious interpretation of single-delay ASL data in pediatric studies.

H3K27M-mutant diffuse midline gliomas (DMGs) express high levels of the disialoganglioside GD2 (ref. ¹). Chimeric antigen receptor-modified T cells targeting GD2 (GD2-CART) eradicated DMGs in preclinical models¹. Arm A of Phase I trial no. NCT04196413 (ref. ²) administered one intravenous (IV) dose of autologous GD2-CART to patients with H3K27M-mutant pontine (DIPG) or spinal DMG (sDMG) at two dose levels (DL1, 1 × 10⁶ kg^{− 1}; DL2, 3 × 10⁶ kg⁻¹) following lymphodepleting chemotherapy. Patients with clinical or imaging benefit were eligible for subsequent intracerebroventricular (ICV) intracranial infusions (10–30 × 10⁶ GD2-CART). Primary objectives were manufacturing feasibility, tolerability and the identification of maximally tolerated IV dose. Secondary objectives included preliminary assessments of benefit. Thirteen patients enroled, with 11 receiving IV GD2-CART on study (n = 3 DL1 (3 DIPG); n = 8 DL2 (6 DIPG, 2 sDMG)). GD2-CART manufacture was successful for all patients. No dose-limiting toxicities occurred on DL1, but three patients experienced dose-limiting cytokine release syndrome on DL2, establishing DL1 as the maximally tolerated IV dose. Nine patients received ICV infusions, with no dose-limiting toxicities. All patients exhibited tumour inflammation-associated neurotoxicity, safely managed with intensive monitoring and care. Four patients demonstrated major volumetric tumour reductions (52, 54, 91 and 100%), with a further three patients exhibiting smaller reductions. One patient exhibited a complete response ongoing for over 30 months since enrolment. Nine patients demonstrated neurological benefit, as measured by a protocol-directed clinical improvement score. Sequential IV, followed by ICV GD2-CART, induced tumour regressions and neurological improvements in patients with DIPG and those with sDMG.

While multiple factors impact disease, artificial intelligence (AI) studies in medicine often use small, non-diverse patient cohorts due to data sharing and privacy issues. Federated learning (FL) has emerged as a solution, enabling training across hospitals without direct data sharing. Here, we present FL-PedBrain, an FL platform for pediatric posterior fossa brain tumors, and evaluate its performance on a diverse, realistic, multi-center cohort. Pediatric brain tumors were targeted due to the scarcity of such datasets, even in tertiary care hospitals. Our platform orchestrates federated training for joint tumor classification and segmentation across 19 international sites. FL-PedBrain exhibits less than a 1.5% decrease in classification and a 3% reduction in segmentation performance compared to centralized data training. FL boosts segmentation performance by 20 to 30% on three external, out-of-network sites. Finally, we explore the sources of data heterogeneity and examine FL robustness in real-world scenarios with data imbalances.

BACKGROUND AND PURPOSE: Molecular biomarker identification increasingly influences the treatment planning of pediatric low-grade neuroepithelial tumors (PLGNTs). We aimed to develop and validate a radiomics-based ADC signature predictive of the molecular status of PLGNTs. MATERIALS AND METHODS: In this retrospective bi-institutional study, we searched the PACS for baseline brain MRIs from children with PLGNTs. Semiautomated tumor segmentation on ADC maps was performed using the semiautomated level tracing effect tool with 3D Slicer. Clinical variables, including age, sex, and tumor location, were collected from chart review. The molecular status of tumors was derived from biopsy. Multiclass random forests were used to predict the molecular status and fine-tuned using a grid search on the validation sets. Models were evaluated using independent and unseen test sets based on the combined data, and the area under the receiver operating characteristic curve (AUC) was calculated for the prediction of 3 classes: KIAA1549-BRAF fusion, BRAF V600E mutation, and non-BRAF cohorts. Experiments were repeated 100 times using different random data splits and model initializations to ensure reproducible results. RESULTS: Two hundred ninety-nine children from the first institution and 23 children from the second institution were included (53.6% male; mean, age 8.01 years; 51.8% supratentorial; 52.2% with KIAA1549-BRAF fusion). For the 3-class prediction using radiomics features only, the average test AUC was 0.74 (95% CI, 0.73–0.75), and using clinical features only, the average test AUC was 0.67 (95% CI, 0.66–0.68). The combination of both radiomics and clinical features improved the AUC to 0.77 (95% CI, 0.75–0.77). The diagnostic performance of the per-class test AUC was higher in identifying KIAA1549-BRAF fusion tumors among the other subgroups (AUC = 0.81 for the combined radiomics and clinical features versus 0.75 and 0.74 for BRAF V600E mutation and non-BRAF, respectively). CONCLUSIONS: ADC values of tumor segmentations have differentiative signals that can be used for training machine learning classifiers for molecular biomarker identification of PLGNTs. ADC-based pretherapeutic differentiation of the BRAF status of PLGNTs has the potential to avoid invasive tumor biopsy and enable earlier initiation of targeted therapy.

The third ML4H symposium was held in person on December 10, 2023, in New Orleans, Louisiana, USA. The symposium included research roundtable sessions to foster discussions between participants and senior researchers on timely and relevant topics for the \ac{ML4H} community. Encouraged by the successful virtual roundtables in the previous year, we organized eleven in-person roundtables and four virtual roundtables at ML4H 2022. The organization of the research roundtables at the conference involved 17 Senior Chairs and 19 Junior Chairs across 11 tables. Each roundtable session included invited senior chairs (with substantial experience in the field), junior chairs (responsible for facilitating the discussion), and attendees from diverse backgrounds with interest in the session's topic. Herein we detail the organization process and compile takeaways from these roundtable discussions, including recent advances, applications, and open challenges for each topic. We conclude with a summary and lessons learned across all roundtables. This document serves as a comprehensive review paper, summarizing the recent advancements in machine learning for healthcare as contributed by foremost researchers in the field.

BACKGROUND AND PURPOSE: MS is a chronic progressive, idiopathic, demyelinating disorder whose diagnosis is contingent on the interpretation of MR imaging. New MR imaging lesions are an early biomarker of disease progression. We aimed to evaluate a machine learning model based on radiomics features in predicting progression on MR imaging of the brain in individuals with MS. MATERIALS AND METHODS: This retrospective cohort study with external validation on open-access data obtained full ethics approval. Longitudinal MR imaging data for patients with MS were collected and processed for machine learning. Radiomics features were extracted at the future location of a new lesion in the patients’ prior MR imaging (“prelesion”). Additionally, “control” samples were obtained from the normal-appearing white matter for each participant. Machine learning models for binary classification were trained and tested and then evaluated the external data of the model. RESULTS: The total number of participants was 167. Of the 147 in the training/test set, 102 were women and 45 were men. The average age was 42 (range, 21–74 years). The best-performing radiomics-based model was XGBoost, with accuracy, precision, recall, and F1-score of 0.91, 0.91, 0.91, and 0.91 on the test set, and 0.74, 0.74, 0.74, and 0.70 on the external validation set. The 5 most important radiomics features to the XGBoost model were associated with the overall heterogeneity and low gray-level emphasis of the segmented regions. Probability maps were produced to illustrate potential future clinical applications. CONCLUSIONS: Our machine learning model based on radiomics features successfully differentiated prelesions from normal-appearing white matter. This outcome suggests that radiomics features from normal-appearing white matter could serve as an imaging biomarker for progression of MS on MR imaging.

Desmoplastic infantile ganglioglioma/desmoplastic infantile astrocytomas (DIG/DIA), although rare, are low-grade glioneuronal tumors that typically affect patients < 1 year old, and generally carry a good prognosis. Often successfully cured with surgical excision, DIG/DIA still carry the potential for malignant transformation or recurrence. Whilst pediatric low grade glioma (LGG) survival rates are >95%, tumor grade does not correlate well with outcomes in infant LGG (iLGG), thus highlighting the need for investigation into molecular mechanisms of tumor progression. We describe a novel CDC42BPA-BRAF fusion in a DIA in a three-month old patient, and its detection via Next Generation Sequencing. Future studies will investigate whether this novel fusion can function as a target for BRAF and MAPK pathway inhibitors, which have recently been FDA-approved for pLGG.