Sanjay P. Prabhu

The BraTS-PEDs dataset provides the largest, multi-institutional, publicly available collection of annotated multiparametric MRI scans of pediatric high-grade gliomas, enabling standardized benchmarking and development of AI tools in pediatric neuro-oncology.

Artificial intelligence applied to brain magnetic resonance imaging (MRI) holds potential to advance diagnosis, prognosis and treatment planning for neurological diseases. The field has been constrained, thus far, by limited training data and task-specific models that do not generalize well across patient populations and medical tasks. By leveraging self-supervised learning, pretraining and targeted adaptation, foundation models present a promising paradigm to overcome these limitations. Here we present Brain Imaging Adaptive Core (BrainIAC)—a foundation model designed to learn generalized representations from unlabeled brain MRI data and serve as a core basis for diverse downstream application adaptation. Trained and validated on 48,965 brain MRIs across a broad spectrum of tasks, we demonstrate that BrainIAC outperforms localized supervised training and other pretrained models, particularly in low-data, few-shot, settings and in high-difficulty prediction tasks, allowing for application in scenarios otherwise infeasible. BrainIAC can be integrated into imaging pipelines and multimodal frameworks and may lead to improved biomarker discovery and artificial intelligence clinical translation.

Pediatric glioma recurrence can cause morbidity and mortality; however, recurrence patterns and severity are heterogeneous and challenging to predict with established clinical and genomic markers. As a result, almost all children undergo frequent, long-term, magnetic resonance imaging (MRI) brain surveillance regardless of individual recurrence risk. Longitudinal deep-learning analysis of serial MRI scans may be an effective approach for improving individualized recurrence prediction in gliomas and other cancers, but, thus far, progress has been limited by data availability and current machine-learning approaches. We developed a self-supervised temporal deep-learning approach tailored for longitudinal medical imaging analysis, wherein a multistep model encodes patients' serial MRI scans and is trained to classify the correct chronological order as a pretext task. The pretrained model is then fine-tuned to predict the primary end point of interest - in this case, 1-year recurrence prediction for pediatric gliomas from the point of last scan - by leveraging a patient's historical postoperative surveillance scans. We apply the model across 3994 scans from 715 patients followed at three separate institutions in the setting of pediatric low- and high-grade gliomas. Longitudinal imaging analysis with temporal learning improved recurrence prediction performance (F1 score) by up to 58.5% (range, 6.6 to 58.5%) compared with traditional approaches across datasets, with performance improvements in both low- and high-grade gliomas and area under the receiver operating characteristic curve of (range, 75 to 89%) across all datasets. Recurrence prediction performance increased incrementally with the number of historical scans available per patient, reaching plateaus between three and six scans, depending on the dataset. Temporal deep learning enables high-performing longitudinal medical imaging analysis and point-of-care decision support for pediatric brain tumors. Temporal learning may be broadly adaptable to track and predict risk in patients with other cancers and chronic diseases undergoing surveillance imaging. (Funded in part by the National Institutes of Health/National Cancer Institute (U54 CA274516 and P50 CA165962), and Botha-Chan Low Grade Glioma Consortium.).

Objective: Sturge–Weber syndrome (SWS) is a neurocutaneous disorder caused by a somatic mosaic mutation in the GNAQ gene. Epilepsy is seen in 75%–80% of children with SWS, and they are at high risk of early onset seizures, status epilepticus, and drug‐resistant epilepsy. Epilepsy surgery is an effective treatment, but timing and candidacy for epilepsy surgery remain controversial in this patient population. We examined the indications, trends, and outcomes of surgically treated patients with SWS at a large‐volume tertiary pediatric hospital over >2 decades. Methods: This retrospective cohort study includes all patients who have a clinical diagnosis of SWS, underwent epilepsy surgery, and were followed by a neurologist at Boston Children's Hospital with clinic visit(s) between January 2000 and April 2022. Chart review and descriptive statistics were performed, and epilepsy characteristics and magnetic resonance imaging findings of the surgical cohort were compared with a nonsurgical SWS cohort. Results: Seventeen patients met the inclusion criteria, 12 with unilateral brain involvement (six right hemispheric) and five with bilateral asymmetric involvement. The average age at seizure onset was 7 months, and average age at first surgery was 29 months. Indication for epilepsy surgery was medically refractory epilepsy in all patients. Eleven hemispherectomies (seven anatomical, four functional) and six motor‐sparing focal resections/disconnections were performed. After an average follow‐up time of 7 years, Engel class I or II seizure outcome was achieved in 15 patients, including all five with bilateral asymmetric involvement. Five patients have discontinued antiseizure medications. Significance: Outcomes after epilepsy surgery in children with SWS are robust and durable, with a low rate of complications. Both functional hemispherectomy and focal resections are effective in carefully selected patients, and patients with bilateral asymmetric involvement may have favorable outcomes.

Somatic mosaic variants contribute to focal epilepsy, with variants often present only in brain tissue and not in blood or other samples typically assayed for genetic testing. Thus, genetic analysis for mosaic variants in focal epilepsy has been limited to patients with drug-resistant epilepsy who undergo surgical resection and have resected brain tissue samples available. Stereo-EEG (sEEG) has become part of the evaluation for many patients with focal drug-resistant epilepsy, and sEEG electrodes provide a potential source of small amounts of brain-derived DNA. We aimed to identify, validate, and assess the distribution of deleterious mosaic variants in epilepsy-associated genes in DNA extracted from trace brain tissue on individual sEEG electrodes. We enrolled a prospective cohort of 10 paediatric patients with drug-resistant epilepsy who had sEEG electrodes implanted for invasive monitoring. We extracted unamplified DNA and in parallel performed whole-genome amplification from trace brain tissue on each sEEG electrode. We also extracted DNA from resected brain tissue and blood/saliva samples where available. We performed deep sequencing (panel and exome) and analysis for candidate germline and mosaic variants. We validated candidate mosaic variants and assessed the variant allele fraction in amplified and unamplified electrode–derived DNA and across electrodes. We extracted unamplified DNA and performed whole-genome amplification from >150 individual electrodes from 10 individuals. Immunohistochemistry confirmed the presence of neurons in the brain tissue on electrodes. Deep sequencing and analysis demonstrated similar depth of coverage between amplified and unamplified DNA samples but significantly more potential mosaic variants in amplified samples. We validated four deleterious mosaic variants in epilepsy-associated genes in electrode-derived DNA in three patients who underwent laser ablation and did not have resected brain tissue samples available. Three of the four variants were detected in both amplified and unamplified electrode–derived DNA, with higher variant allele fraction observed in DNA from electrodes in closest proximity to the electrical seizure focus in one case. We demonstrate that mosaic variants can be identified and validated from DNA extracted from trace brain tissue on individual sEEG electrodes in patients with drug-resistant focal epilepsy, from both unamplified and amplified electrode–derived DNA. Our findings support a relationship between the extent of regional genetic abnormality and electrophysiology and suggest that with further optimization, this minimally invasive diagnostic approach holds promise for advancing precision medicine for patients with drug-resistant epilepsy as part of the surgical evaluation.

Sensory integration is the province of the parietal lobe. The non-dominant hemisphere is responsible for both body sides, while the dominant hemisphere is responsible for the contralateral hemi-body. Furthermore, the posterior cingulate cortex (PCC) participates in a network involved in spatial orientation, attention, and spatial and episodic memory. Laser interstitial thermotherapy (LiTT) is a minimally invasive surgery for focal drug-resistant epilepsy (DRE) that can target deeper brain regions, and thus, region-specific symptoms can emerge. Here, we present an 18-year-old right-handed male with focal DRE who experienced seizures characterized by sensations of déjà vu, staring spells, and language disruption. A comprehensive evaluation localized the seizure focus and revealed a probable focal cortical dysplasia (FCD) in the left posterior cingulate gyrus. The patient underwent uneventful LiTT of the identified lesion. Post-operatively, he developed transient ipsilateral spatial neglect and contralateral sensory loss, as well as acalculia. His sensory symptoms gradually improved after the surgery, and he remained seizure-free after the intervention for at least 10 months (until the time of this writing). This rare case of ipsilateral spatial and visual hemineglect post-LiTT in epilepsy underscores the importance of recognizing atypical neurosurgical outcomes and considering individual variations in brain anatomy and function. Understanding the dynamics of cortical connectivity and handedness, particularly in pediatric epilepsy, may be crucial in anticipating and managing neurocognitive effects following epilepsy surgery.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To develop, externally test, and evaluate clinical acceptability of a deep learning (DL) pediatric brain tumor segmentation model using stepwise transfer learning. Materials and Methods In this retrospective study, the authors leveraged two T2-weighted MRI datasets (May 2001-December 2015) from a national brain tumor consortium (n = 184; median age, 7 years (range: 1-23 years); 94 male) and a pediatric cancer center (n = 100; median age, 8 years (range: 1-19 years); 47 male) to develop and evaluate DL neural networks for pediatric low-grade glioma segmentation using a novel stepwise transfer learning approach to maximize performance in a limited data scenario. The best model was externally-tested on an independent test set and subjected to randomized, blinded evaluation by three clinicians, wherein they assessed clinical acceptability of expert- and artificial intelligence (AI)-generated segmentations via 10-point Likert scales and Turing tests. Results The best AI model used in-domain, stepwise transfer learning (median DSC: 0.88 [IQR 0.72-0.91] versus 0.812 [0.56-0.89] for baseline model; P = .049). On external testing, AI model yielded excellent accuracy using reference standards from three clinical experts (Expert-1: 0.83 [0.75-0.90]; Expert-2: 0.81 [0.70-0.89]; Expert-3: 0.81 [0.68-0.88]; mean accuracy: 0.82)). On clinical benchmarking (n = 100 scans), experts rated AI-based segmentations higher on average compared with other experts (median Likert score: median 9 [IQR 7-9]) versus 7 [IQR 7-9]) and rated more AI segmentations as clinically acceptable (80.2% versus 65.4%). Experts correctly predicted the origin of AI segmentations in an average of 26.0% of cases. Conclusion Stepwise transfer learning enabled expert-level, automated pediatric brain tumor auto-segmentation and volumetric measurement with a high level of clinical acceptability. ©RSNA, 2024.