SignificancePredicting the expected outcome of patients diagnosed with cancer is a critical step in treatment. Advances in genomic and imaging technologies provide physicians with vast amounts of data, yet prognostication remains largely subjective, leading to suboptimal clinical management. We developed a computational approach based on deep learning to predict the overall survival of patients diagnosed with brain tumors from microscopic images of tissue biopsies and genomic biomarkers. This method uses adaptive feedback to simultaneously learn the visual patterns and molecular biomarkers associated with patient outcomes. Our approach surpasses the prognostic accuracy of human experts using the current clinical standard for classifying brain tumors and presents an innovative approach for objective, accurate, and integrated prediction of patient outcomes.
Cancer histology reflects underlying molecular processes and disease progression, and contains rich phenotypic information that is predictive of patient outcomes. In this study, we demonstrate a computational approach for learning patient outcomes from digital pathology images using deep learning to combine the power of adaptive machine learning algorithms with traditional survival models. We illustrate how this approach can integrate information from both histology images and genomic biomarkers to predict time-to-event patient outcomes, and demonstrate performance surpassing the current clinical paradigm for predicting the survival of patients diagnosed with glioma. We also provide techniques to visualize the tissue patterns learned by these deep learning survival models, and establish a framework for addressing intratumoral heterogeneity and training data deficits.
Translating the vast data generated by genomic platforms into accurate predictions of clinical outcomes is a fundamental challenge in genomic medicine. Many prediction methods face limitations in learning from the high-dimensional profiles generated by these platforms, and rely on experts to hand-select a small number of features for training prediction models. In this paper, we demonstrate how deep learning and Bayesian optimization methods that have been remarkably successful in general high-dimensional prediction tasks can be adapted to the problem of predicting cancer outcomes. We perform an extensive comparison of Bayesian optimized deep survival models and other state of the art machine learning methods for survival analysis, and describe a framework for interpreting deep survival models using a risk backpropagation technique. Finally, we illustrate that deep survival models can successfully transfer information across diseases to improve prognostic accuracy. We provide an open-source software implementation of this framework called SurvivalNet that enables automatic training, evaluation and interpretation of deep survival models.
Glioblastoma (GBM) is the most aggressive malignant primary brain tumor in adults, with a median survival of 14.6 months. Recent efforts have focused on identifying clinically relevant subgroups to improve our understanding of pathogenetic mechanisms and patient stratification. Concurrently, the role of immune cells in the tumor microenvironment has received increasing attention, especially T cells and tumor-associated macrophages (TAM). The latter are a mixed population of activated brain-resident microglia and infiltrating monocytes/monocyte-derived macrophages, both of which express ionized calcium-binding adapter molecule 1 (IBA1). This study investigated differences in immune cell subpopulations among distinct transcriptional subtypes of GBM. Human GBM samples were molecularly characterized and assigned to Proneural, Mesenchymal or Classical subtypes as defined by NanoString nCounter Technology. Subsequently, we performed and analyzed automated immunohistochemical stainings for TAM as well as specific T cell populations. The Mesenchymal subtype of GBM showed the highest presence of TAM, CD8 + , CD3 + and FOXP3 + T cells, as compared to Proneural and Classical subtypes. High expression levels of the TAM-related gene AIF1, which encodes the TAM-specific protein IBA1, correlated with a worse prognosis in Proneural GBM, but conferred a survival benefit in Mesenchymal tumors. We used our data to construct a mathematical model that could reliably identify Mesenchymal GBM with high sensitivity using a combination of the aforementioned cell-specific IHC markers. In conclusion, we demonstrated that molecularly distinct GBM subtypes are characterized by profound differences in the composition of their immune microenvironment, which could potentially help to identify tumors amenable to immunotherapy.
Objective:We assessed whether peripheral activation of microglia by a nasal proteosome-based adjuvant (Protollin) that has been given safely to humans can prevent amyloid deposition in young mice and affect amyloid deposition and memory function in old mice with a large amyloid load. Methods: Amyloid precursor protein (APP) transgenic (Tg) J20 mice received nasal treatment with Protollin weekly for 8 months beginning at age 5 months. Twenty-four-month-old J20 mice were treated weekly for 6 weeks. Results: We found reduction in the level of fibrillar amyloid (93%), insoluble -amyloid (A; 68%), and soluble A (45%) fragments in 14-month-old mice treated with Protollin beginning at age 5 months. Twenty-four-month-old mice treated with nasal Protollin for 6 weeks had decreased soluble and insoluble A (1-40) and (1-42) and improved memory function. Activated microglia (CD11b ϩ cells) colocalized with A fibrils in the 24-month-old animals, and microglial activation correlated with the decrease in A. No microglial activation was observed in 14-month-old mice, suggesting that once A is cleared, there is downregulation of microglial activation. Both groups had reduction in astrocytosis. Protollin was observed in the nasal cavity and cervical lymph node but not in the brain. Activated CD11b ϩ SRA ϩ (scavenger receptor A) cells were found in blood and cervical lymph node and increased interleukin-10 in cervical lymph node. No toxicity was associated with treatment.Interpretation: Our results demonstrate a novel antibody-independent immunotherapy for both prevention and treatment of Alzheimer's disease that is mediated by peripheral activation of microglia with no apparent toxicity.
Meningeal solitary fibrous tumor (SFT)/hemangiopericytoma (HPC) is a rare tumor with propensity for recurrence and metastasis. Although multiple classification schemes have been proposed, optimal risk stratification remains unclear, and the prognostic impact of fusion status is uncertain. We compared the 2016 WHO CNS tumor grading scheme (CNS-G), a three-tier system based on histopathologic phenotype and mitotic count, to the 2013 WHO soft-tissue counterpart (ST-G), a two-tier system based on mitotic count alone, in a cohort of 133 patients [59 female, 74 male; mean age 54 years (range 20–87)] with meningeal SFT/HPC. Tumors were pathologically confirmed through review of the first tumor resection ( n = 97), local recurrence ( n = 35), or distant metastasis ( n = 1). A STAT6 immunostain showed nuclear expression in 132 cases. NAB2 – STAT6 fusion was detected in 99 of 111 successfully tested tumors (89%) including the single STAT6 immunonegative tumor. Tumors were classified by CNS-G as grade 1 ( n = 43), 2 ( n = 41), or 3 ( n = 49), and by ST-G as SFT ( n = 84) or malignant SFT ( n = 49). Necrosis was present in 16 cases (12%). On follow-up, 42 patients had at least one subsequent recurrence or metastasis (7 metastasis only, 33 recurrence only, 2 patients had both). Twenty-nine patients died. On univariate analysis, necrosis ( p = 0.002), CNS-G ( p = 0.01), and ST-G ( p = 0.004) were associated with recurrence-free (RFS) but not overall survival (OS). NAB2 – STAT6 fusion type was not significantly associated with RFS or OS, but was associated with phenotype. A modified ST-G incorporating necrosis showed higher correlation with RFS ( p = 0.0006) and remained significant ( p = 0.02) when considering only the primary tumors. From our data, mitotic rate and necrosis appear to stratify this family of tumors most accurately and could be incorporated in a future grading scheme. Electronic supplementary material The online version of this article (10.1007/s00401-018-1952-6) contains supplementary material, which is available to authorized users.
Brain tumors are the leading cause of cancer-related mortality in children and have distinct genomic and molecular features compared with adult glioma. However, the properties of immune cells in these tumors has been vastly understudied compared with their adult counterparts. We combined multiplex immunofluorescence immunohistochemistry coupled with machine learning and single-cell mass cytometry to evaluate T-cells infiltrating pediatric glial tumors. We show that low-grade tumors are characterized by greater T-cell density compared with high-grade glioma (HGG). However, even among low-grade tumors, T-cell infiltration can be highly variable and subtype-dependent, with greater T-cell density in pleomorphic xanthoastrocytoma and ganglioglioma. CD3+ T-cell infiltration correlates inversely with the expression of SOX2, an embryonal stem cell marker commonly expressed by glial tumors. T-cells within both HGG and low-grade glioma (LGG) exhibit phenotypic heterogeneity and tissue-resident memory T-cells consist of distinct subsets of CD103+ and TCF1+ cells that exhibit distinct spatial localization patterns. TCF1+ T-cells are located closer to the vessels while CD103+ resident T-cells reside within the tumor further away from the vasculature. Recurrent tumors are characterized by a decline in CD103+ tumor-infiltrating T-cells. BRAFV600E mutation is immunogenic in children with LGG and may serve as a target for immune therapy. These data provide several novel insights into the subtype-dependent and grade-dependent changes in immune architecture in pediatric gliomas and suggest that harnessing tumor-resident T-cells may be essential to improve immune control in glioma.
DICER1 mutations (somatic or germline) are associated with a variety of uncommon neoplasms including cervical and genitourinary embryonal rhabdomyosarcoma (ERMS). We report a primary ovarian and 2 primary fallopian tube ERMS occurring in 60-, 13-, and 14-year-olds, respectively. The 3 neoplasms exhibited a similar morphologic appearance being polypoid and containing edematous hypocellular areas and hypercellular foci composed of small cells with scant cytoplasm exhibiting rhabdomyoblastic differentiation (desmin, myogenin, myoD1 positive). There was cellular cartilage in all cases and extensive foci of anaplasia, eosinophilic globules, and bone/osteoid in 1 case each. All 3 neoplasms exhibited DICER1 mutations; in 1 of the tubal cases, the patient had a germline mutation and in the other 2 cases, the DICER1 mutations were somatic. Accompanying DICER1 “second hits” were identified in all cases. In 2 of the neoplasms, SALL4-positive glandular structures were present which we speculate may represent an unusual primitive “metaplastic” phenomenon. Our study adds to the literature on ERMS at unusual sites associated with DICER1 mutations. ERMS arising at such sites, especially when they contain cartilage or bone/osteoid, are especially likely to be associated with DICER1 mutations. Pathologists should be aware of this as these may be the sentinel neoplasms in patients with DICER1 syndrome and confirming a germline mutation can facilitate the screening of the individual and affected family members for other neoplasms which occur in this syndrome.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.