Search

Gliomas are the most frequent primary brain tumors in adults. Glioma change detection aims at finding the relevant parts of the image that change over time. Although Deep Learning (DL) shows promising performances in similar change detection tasks, the creation of large annotated datasets represents a major bottleneck for supervised DL applications in radiology. To overcome this, we propose a combined use of weak labels (imprecise, but fast-to-create annotations) and Transfer Learning (TL). Specifically, we explore inductive TL, where source and target domains are identical, but tasks are different due to a label shift: our target labels are created manually by three radiologists, whereas our source weak labels are generated automatically from radiology reports via NLP. We frame knowledge transfer as hyperparameter optimization, thus avoiding heuristic choices that are frequent in related works. We investigate the relationship between model size and TL, comparing a low-capacity VGG with a higher-capacity ResNeXt model. We evaluate our models on 1693 T2-weighted magnetic resonance imaging difference maps created from 183 patients, by classifying them into stable or unstable according to tumor evolution. The weak labels extracted from radiology reports allowed us to increase dataset size more than 3-fold, and improve VGG classification results from 75% to 82% AUC. Mixed training from scratch led to higher performance than fine-tuning or feature extraction. To assess generalizability, we ran inference on an open dataset (BraTS-2015: 15 patients, 51 difference maps), reaching up to 76% AUC. Overall, results suggest that medical imaging problems may benefit from smaller models and different TL strategies with respect to computer vision datasets, and that report-generated weak labels are effective in improving model performances. Code, in-house dataset and BraTS labels are released., Comment: This work has been submitted as Original Paper to a Journal

Additional file. 1. Table S1.Description of full EORTC/NCIC & LN-Pilot and Nordic datasets. Table S2. DNA methylation associated with patient age. Table S3. Associations of methylation based entropy by genomic region and GBM classification. Table S4. Functional CpGs located on DDR genes, associated with Age Acceleration. Table S5. Functional CpGs located on DDR genes, associated with GBM classification. Fig. S1 Methylation-based classification of tumors of all patients from EORTC/NCIC & LN-Pilot and Nordic studies. Fig. S2 Impact of sample purity on measures of DNAm age acceleration, and HM entropy. Fig. S3 Pathways associated with functional methylation related to DNAm age acceleration and GBM classification. Fig. S4 CDKN2A CNV and DNAm age acceleration.

Contains fulltext : 248925.pdf (Publisher’s version ) (Open Access)

Additional file 1. Fig. S1 Flow charts for gene selection procedure. Fig. S2. Longitudinal variations of metabolite concentrations of PDOX. Fig. S3. Multivoxel 1H MRS at 7 Tesla of patient 1. Fig. S4. Retention of characteristic molecular features in PDOX. Fig. S5. Expression analyses including all PDOX and original human GBMs.

The use of methylated DNA capturing followed by massive parallel sequencing has become a robust method for the study of genome-wide methylation patterns. However, it has recently been demonstrated that the accurate estimation of methylation levels is dependent on incorporating CNV status of the sampled genomic regions (Robinson et al, 2012). We used Illumina multiplex paired-end (PE) sequencing of MBD-captured DNA isolated from GBM samples. The mean number of PE reads per library was 5.3 million [19.3-2 million]. Alignment of quality-controlled (min. Q20) and de-duplicated reads to the human reference genome hg19 was performed using bowtie2, resulting in a mean number of mapped PE reads of 3.8 million [14-1 million]. Our preliminary results indicate differences in the methylation of CpG islands located in promoter regions when comparing normal versus tumour samples. We also performed DNA methylation estimation using the Infinium 450k platform for the same set of samples. Illumina's Infinium HumanMethylation450 technology is based on the OmniBeadChip genotyping technology which has been used to accurately estimate CNVs. We therefore hypothesized that the 450k methylation data (signal intensities) could be used for the estimation of the CNV status. One pre-requisite for this approach is the availability of a canonical set of controls generated from normal tissue. Due to the limited sample size of our data set, we plan to incorporate the 450k methylation and CNV data available from TCGA to: 1. create the canonical set of controls of 450k data from normal tissues of TCGA samples 2. perform CNV estimation of TCGA samples based on 450k methylation data 3. verify the accuracy of the 450k-based approach against TCGA CNV data The use of TCGA data in this methodology will allow our laboratory to evaluate the robustness of this approach before applying it to our own datasets.

For patients with brain tumors identification of diagnostic and prognostic markers in easy accessible biological material, such as plasma or cerebrospinal fluid (CSF), would greatly facilitate patient management. MIC-1/GDF15 (growth differentiation factor 15) is a secreted protein of the TGF-beta superfamily and emerged as a candidate marker exhibiting increasing mRNA expression during malignant progression of glioma. Determination of MIC-1/GDF15 protein levels by ELISA in the CSF of a cohort of 94 patients with intracranial tumors including gliomas, meningioma and metastasis revealed significantly increased concentrations in glioblastoma patients (median, 229 pg/ml) when compared with control cohort of patients treated for non-neoplastic diseases (median below limit of detection of 156 pg/ml, p < 0.0001, Mann-Whitney test). However, plasma MIC-1/GDF15 levels were not elevated in the matching plasma samples from these patients. Most interestingly, patients with glioblastoma and increased CSF MIC-1/GDF15 had a shorter survival (p = 0.007, log-rank test). In conclusion, MIC-1/GDF15 protein measured in the CSF may have diagnostic and prognostic value in patients with intracranial tumors. (c) 2009 UICC

Background: Radiotherapy is widely used to treat cancer. While rapidly dividing cancer cells are naturally considered the main target of radiotherapy, emerging evidence indicates that radiotherapy also affects endothelial cell functions, and possibly also their angiogenic capacity. In spite of its clinical relevance, such putative anti-angiogenic effect of radiotherapy has not been thoroughly characterized. We have investigated the effect of ionizing radiation on angiogenesis using in vivo, ex vivo and in vitro experimental models in combination with genetic and pharmacological interventions.

Glioblastoma (GBM) is the most aggressive brain tumor type in adults. GBM is heterogeneous, with a compact core lesion surrounded by an invasive tumor front. This front is highly relevant for tumor recurrence but is generally non-detectable using standard imaging techniques. Recent studies demonstrated distinct metabolic profiles of the invasive phenotype in GBM. Magnetic resonance (MR) of hyperpolarized C-13-labeled probes is a rapidly advancing field that provides real-time metabolic information. Here, we applied hyperpolarized C-13-glucose MR to mouse GBM models. Compared to controls, the amount of lactate produced from hyperpolarized glucose was higher in the compact GBM model, consistent with the accepted "Warburg effect". However, the opposite response was observed in models reflecting the invasive zone, with less lactate produced than in controls, implying a reduction in aerobic glycolysis. These striking differences could be used to map the metabolic heterogeneity in GBM and to visualize the infiltrative front of GBM.

Purpose: Quantitative methylation-specific tests suggest that not all cells in a glioblastoma with detectable promoter methylation of the O6-methylguanine DNA methyltransferase (MGMT) gene carry a methylated MGMT allele. This observation may indicate cell subpopulations with distinct MGMT status, raising the question of the clinically relevant cutoff of MGMT methylation therapy. Epigenetic silencing of the MGMT gene by promoter methylation blunts repair of O6-methyl guanine and has been shown to be a predictive factor for benefit from alkylating agent therapy in glioblastoma.

Two cost-efficient genome-scale methodologies to assess DNA-methylation are MethylCap-seq and Illumina’s Infinium HumanMethylation450 BeadChips (HM450). Objective information regarding the best-suited methodology for a specific research question is scant. Therefore, we performed a large-scale evaluation on a set of 70 brain tissue samples, i.e. 65 glioblastoma and 5 non-tumoral tissues. As MethylCap-seq coverages were limited, we focused on the inherent capacity of the methodology to detect methylated loci rather than a quantitative analysis. MethylCap-seq and HM450 data were dichotomized and performances were compared using a gold standard free Bayesian modelling procedure. While conditional specificity was adequate for both approaches, conditional sensitivity was systematically higher for HM450. In addition, genome-wide characteristics were compared, revealing that HM450 probes identified substantially fewer regions compared to MethylCap-seq. Although results indicated that the latter method can detect more potentially relevant DNA-methylation, this did not translate into the discovery of more differentially methylated loci between tumours and controls compared to HM450. Our results therefore indicate that both methodologies are complementary, with a higher sensitivity for HM450 and a far larger genome-wide coverage for MethylCap-seq, but also that a more comprehensive character does not automatically imply more significant results in biomarker studies.