Summary Accurate pathological diagnosis is crucial for optimal management of cancer patients. For the ~100 known central nervous system (CNS) tumour entities, standardization of the diagnostic process has been shown to be particularly challenging - with substantial inter-observer variability in the histopathological diagnosis of many tumour types. We herein present the development of a comprehensive approach for DNA methylation-based CNS tumour classification across all entities and age groups, and demonstrate its application in a routine diagnostic setting. We show that availability of this method may have substantial impact on diagnostic precision compared with standard methods, resulting in a change of diagnosis in up to 12% of prospective cases. For broader accessibility we have designed a free online classifier tool (www.molecularneuropathology.org) requiring no additional onsite data processing. Our results provide a blueprint for the generation of machine learning-based tumour classifiers across other cancer entities, with the potential to fundamentally transform tumour pathology.

SummaryWe collated data from 157 unpublished cases of pediatric high-grade glioma and diffuse intrinsic pontine glioma and 20 publicly available datasets in an integrated analysis of >1,000 cases. We identified co-segregating mutations in histone-mutant subgroups including loss of FBXW7 in H3.3G34R/V, TOP3A rearrangements in H3.3K27M, and BCOR mutations in H3.1K27M. Histone wild-type subgroups are refined by the presence of key oncogenic events or methylation profiles more closely resembling lower-grade tumors. Genomic aberrations increase with age, highlighting the infant population as biologically and clinically distinct. Uncommon pathway dysregulation is seen in small subsets of tumors, further defining the molecular diversity of the disease, opening up avenues for biological study and providing a basis for functionally defined future treatment stratification.

Isocitrate dehydrogenase () mutations in glioma patients confer longer survival and may guide treatment decision making. We aimed to predict the status of gliomas from MR imaging by applying a residual convolutional neural network to preoperative radiographic data. Preoperative imaging was acquired for 201 patients from the Hospital of University of Pennsylvania (HUP), 157 patients from Brigham and Women's Hospital (BWH), and 138 patients from The Cancer Imaging Archive (TCIA) and divided into training, validation, and testing sets. We trained a residual convolutional neural network for each MR sequence (FLAIR, T2, T1 precontrast, and T1 postcontrast) and built a predictive model from the outputs. To increase the size of the training set and prevent overfitting, we augmented the training set images by introducing random rotations, translations, flips, shearing, and zooming. With our neural network model, we achieved IDH prediction accuracies of 82.8% (AUC = 0.90), 83.0% (AUC = 0.93), and 85.7% (AUC = 0.94) within training, validation, and testing sets, respectively. When age at diagnosis was incorporated into the model, the training, validation, and testing accuracies increased to 87.3% (AUC = 0.93), 87.6% (AUC = 0.95), and 89.1% (AUC = 0.95), respectively. We developed a deep learning technique to noninvasively predict genotype in grade II-IV glioma using conventional MR imaging using a multi-institutional data set. .

Infant high grade gliomas appear clinically distinct from their counterparts in older children, indicating that histopathologic grading may not accurately reflect the biology of these tumors. We have collected 241 cases under 4 years of age, and carried out histological review, methylation profiling, custom panel and genome/exome sequencing. After excluding tumors representing other established entities or subgroups, we identified 130 cases to be part of an 'intrinsic' spectrum of disease specific to the infant population. These included those with targetable MAP-kinase alterations, and a large proportion of remaining cases harboring gene fusions targeting ALK (n=31), NTRK1/2/3 (n=21), ROS1 (n=9) and MET (n=4) as their driving alterations, with evidence of efficacy of targeted agents in the clinic. These data strongly supports the concept that infant gliomas require a change in diagnostic practice and management.

Sudden unexplained death in infants, including the sudden infant death syndrome, is likely due to heterogeneous causes that involve different intrinsic vulnerabilities and/or environmental factors. Neuropathologic research focuses upon the role of brain regions, particularly the brainstem, that regulate or modulate autonomic and respiratory control during sleep or transitions to waking. The hippocampus is a key component of the forebrain–limbic network that modulates autonomic/respiratory control via brainstem connections, but its role in sudden infant death has received little attention. We tested the hypothesis that a well-established marker of hippocampal pathology in temporal lobe epilepsy—focal granule cell bilamination in the dentate, a variant of granule cell dispersion—is associated with sudden unexplained death in infants. In a blinded study of hippocampal morphology in 153 infants with sudden and unexpected death autopsied in the San Diego County medical examiner’s office, deaths were classified as unexplained or explained based upon autopsy and scene investigation. Focal granule cell bilamination was present in 41.2 % (47/114) of the unexplained group compared to 7.7 % (3/39) of the explained (control) group (p < 0.001). It was associated with a cluster of other dentate developmental abnormalities that reflect defective neuronal proliferation, migration, and/or survival. Dentate lesions in a large subset of infants with sudden unexplained death may represent a developmental vulnerability that leads to autonomic/respiratory instability or autonomic seizures, and sleep-related death when the infants are challenged with homeostatic stressors. Importantly, these lesions can be recognized in microscopic sections prepared in current forensic practice. Future research is needed to determine the relationship between hippocampal and previously reported brainstem pathology in sudden infant death.Electronic supplementary materialThe online version of this article (doi:10.1007/s00401-014-1357-0) contains supplementary material, which is available to authorized users.