Convolutional neural networks to predict brain tumor grades and Alzheimer's disease with MR spectroscopic imaging data.

PURPOSE: To evaluate the value of convolutional neural network (CNN) in the diagnosis of human brain tumor or Alzheimer's disease by MR spectroscopic imaging (MRSI) and to compare its Matthews correlation coefficient (MCC) score against that of other machine learning methods and previous evaluation of the same data. We address two challenges: 1) limited number of cases in MRSI datasets and 2) interpretability of results in the form of relevant spectral regions. METHODS: A shallow CNN with only one hidden layer and an ad-hoc loss function was constructed involving two branches for processing spectral and image features of a brain voxel respectively. Each branch consists of a single convolutional hidden layer. The output of the two convolutional layers is merged and fed to a classification layer that outputs class predictions for the given brain voxel. RESULTS: Our CNN method separated glioma grades 3 and 4 and identified Alzheimer's disease patients using MRSI and complementary MRI data with high MCC score (Area Under the Curve were 0.87 and 0.91 respectively). The results demonstrated superior effectiveness over other popular methods as Partial Least Squares or Support Vector Machines. Also, our method automatically identified the spectral regions most important in the diagnosis process and we show that these are in good agreement with existing biomarkers from the literature. CONCLUSION: Shallow CNNs models integrating image and spectral features improved quantitative and exploration and diagnosis of brain diseases for research and clinical purposes. Software is available at https://bitbucket.org/TeslaH2O/cnn_mrsi.

A methodological approach to correlate tumor heterogeneity with drug distribution profile in mass spectrometry imaging data.

BACKGROUND: Drug mass spectrometry imaging (MSI) data contain knowledge about drug and several other molecular ions present in a biological sample. However, a proper approach to fully explore the potential of such type of data is still missing. Therefore, a computational pipeline that combines different spatial and non-spatial methods is proposed to link the observed drug distribution profile with tumor heterogeneity in solid tumor. Our data analysis steps include pre-processing of MSI data, cluster analysis, drug local indicators of spatial association (LISA) map, and ions selection. RESULTS: The number of clusters identified from different tumor tissues. The spatial homogeneity of the individual cluster was measured using a modified version of our drug homogeneity method. The clustered image and drug LISA map were simultaneously analyzed to link identified clusters with observed drug distribution profile. Finally, ions selection was performed using the spatially aware method. CONCLUSIONS: In this paper, we have shown an approach to correlate the drug distribution with spatial heterogeneity in untargeted MSI data. Our approach is freely available in an R package 'CorrDrugTumorMSI'.

Novel data analysis method for multicolour flow cytometry links variability of multiple markers on single cells to a clinical phenotype.

Multicolour Flow Cytometry (MFC) produces multidimensional analytical data on the quantitative expression of multiple markers on single cells. This data contains invaluable biomedical information on (1) the marker expressions per cell, (2) the variation in such expression across cells, (3) the variability of cell marker expression across samples that (4) may vary systematically between cells collected from donors and patients. Current conventional and even advanced data analysis methods for MFC data explore only a subset of these levels. The Discriminant Analysis of MultiAspect CYtometry (DAMACY) we present here provides a comprehensive view on health and disease responses by integrating all four levels. We validate DAMACY by using three distinct datasets: in vivo response of neutrophils evoked by systemic endotoxin challenge, the clonal response of leukocytes in bone marrow of acute myeloid leukaemia (AML) patients, and the complex immune response in blood of asthmatics. DAMACY provided good accuracy 91-100% in the discrimination between health and disease, on par with literature values. Additionally, the method provides figures that give insight into the marker expression and cell variability for more in-depth interpretation, that can benefit both physicians and biomedical researchers to better diagnose and monitor diseases that are reflected by changes in blood leukocytes.

Lactate and glycine-potential MR biomarkers of prognosis in estrogen receptor-positive breast cancers.

Breast cancer is a heterogeneous disease with a variable prognosis. Clinical factors provide some information about the prognosis of patients with breast cancer; however, there is a need for additional information to stratify patients for improved and more individualized treatment. The aim of this study was to examine the relationship between the metabolite profiles of breast cancer tissue and 5-year survival. Biopsies from breast cancer patients (n=98) were excised during surgery and analyzed by high-resolution magic angle spinning MRS. The data were analyzed by multivariate principal component analysis and partial least-squares discriminant analysis, and the findings of important metabolites were confirmed by spectral integration of the metabolite peaks. Predictions of 5-year survival using metabolite profiles were compared with predictions using clinical parameters. Based on the metabolite profiles, patients with estrogen receptor (ER)-positive breast cancer (n=71) were separated into two groups with significantly different survival rates (p=0.024). Higher levels of glycine and lactate were found to be associated with lower survival rates by both multivariate analyses and spectral integration, and are suggested as biomarkers for breast cancer prognosis. Similar metabolic differences were not observed for ER-negative patients, where survivors could not be separated from nonsurvivors. Predictions of 5-year survival of ER-positive patients using metabolite profiles gave better and more robust results than those using traditional clinical parameters. The results imply that the metabolic state of a tumor may provide additional information concerning breast cancer prognosis. Further studies should be conducted in order to evaluate the role of MR metabolomics as an additional clinical tool for determining the prognosis of patients with breast cancer.

A phase and frequency alignment protocol for 1H MRSI data of the prostate.

(1)H MRSI of the prostate reveals relative metabolite levels that vary according to the presence or absence of tumour, providing a sensitive method for the identification of patients with cancer. Current interpretations of prostate data rely on quantification algorithms that fit model metabolite resonances to individual voxel spectra and calculate relative levels of metabolites, such as choline, creatine, citrate and polyamines. Statistical pattern recognition techniques can potentially improve the detection of prostate cancer, but these analyses are hampered by artefacts and sources of noise in the data, such as variations in phase and frequency of resonances. Phase and frequency variations may arise as a result of spatial field gradients or local physiological conditions affecting the frequency of resonances, in particular those of citrate. Thus, there are unique challenges in developing a peak alignment algorithm for these data. We have developed a frequency and phase correction algorithm for automatic alignment of the resonances in prostate MRSI spectra. We demonstrate, with a simulated dataset, that alignment can be achieved to a phase standard deviation of 0.095 rad and a frequency standard deviation of 0.68 Hz for the citrate resonances. Three parameters were used to assess the improvement in peak alignment in the MRSI data of five patients: the percentage of variance in all MRSI spectra explained by their first principal component; the signal-to-noise ratio of a spectrum formed by taking the median value of the entire set at each spectral point; and the mean cross-correlation between all pairs of spectra. These parameters showed a greater similarity between spectra in all five datasets and the simulated data, demonstrating improved alignment for phase and frequency in these spectra. This peak alignment program is expected to improve pattern recognition significantly, enabling accurate detection and localisation of prostate cancer with MRSI.

The impact of delayed storage on the measured proteome and metabolome of human cerebrospinal fluid.

BACKGROUND: Because cerebrospinal fluid (CSF) is in close contact with diseased areas in neurological disorders, it is an important source of material in the search for molecular biomarkers. However, sample handling for CSF collected from patients in a clinical setting might not always be adequate for use in proteomics and metabolomics studies. METHODS: We left CSF for 0, 30, and 120 min at room temperature immediately after sample collection and centrifugation/removal of cells. At 2 laboratories CSF proteomes were subjected to tryptic digestion and analyzed by use of nano-liquid chromatography (LC) Orbitrap mass spectrometry (MS) and chipLC quadrupole TOF-MS. Metabolome analysis was performed at 3 laboratories by NMR, GC-MS, and LC-MS. Targeted analyses of cystatin C and albumin were performed by LC-tandem MS in the selected reaction monitoring mode. RESULTS: We did not find significant changes in the measured proteome and metabolome of CSF stored at room temperature after centrifugation, except for 2 peptides and 1 metabolite, 2,3,4-trihydroxybutanoic (threonic) acid, of 5780 identified peptides and 93 identified metabolites. A sensitive protein stability marker, cystatin C, was not affected. CONCLUSIONS: The measured proteome and metabolome of centrifuged human CSF is stable at room temperature for up to 2 hours. We cannot exclude, however, that changes undetectable with our current methodology, such as denaturation or proteolysis, might occur because of sample handling conditions. The stability we observed gives laboratory personnel at the collection site sufficient time to aliquot samples before freezing and storage at -80 °C.

On the relevance of automatically selected single-voxel MRS and multimodal MRI and MRSI features for brain tumour differentiation.

In order to evaluate the relevance of magnetic resonance (MR) features selected by automatic feature selection techniques to build classifiers for differential diagnosis and tissue segmentation two data sets containing MR spectroscopy data from patients with brain tumours were investigated. The automatically selected features were evaluated using literature and clinical experience. It was observed that a significant part of the automatically selected features correspond to what is known from the literature and clinical experience. We conclude that automatic feature selection is a useful tool to obtain relevant and possibly interesting features, but evaluation of the obtained features remains necessary.

Alignment of high resolution magic angle spinning magnetic resonance spectra using warping methods.

The peaks of magnetic resonance (MR) spectra can be shifted due to variations in physiological and experimental conditions, and correcting for misaligned peaks is an important part of data processing prior to multivariate analysis. In this paper, five warping algorithms (icoshift, COW, fastpa, VPdtw and PTW) are compared for their feasibility in aligning spectral peaks in three sets of high resolution magic angle spinning (HR-MAS) MR spectra with different degrees of misalignments, and their merits are discussed. In addition, extraction of information that might be present in the shifts is examined, both for simulated data and the real MR spectra. The generic evaluation methodology employs a number of frequently used quality criteria for evaluation of the alignments, together with PLS-DA to assess the influence of alignment on the classification outcome. Peak alignment greatly improved the internal similarity of the data sets. Especially icoshift and COW seem suitable for aligning HR-MAS MR spectra, possibly because they perform alignment segment-wise. The choice of reference spectrum can influence the alignment result, and it is advisable to test several references. Information from the peak shifts was extracted, and in one case cancer samples were successfully discriminated from normal tissue based on shift information only. Based on these findings, general recommendations for alignment of HR-MAS MRS data are presented. Where possible, observations are generalized to other data types (e.g. chromatographic data).

Triple-negative breast cancer: present challenges and new perspectives.

Triple-negative breast cancers (TNBC), characterized by absence of estrogen receptor (ER), progesterone receptor (PR) and lack of overexpression of human epidermal growth factor receptor 2 (HER2), are typically associated with poor prognosis, due to aggressive tumor phenotype(s), only partial response to chemotherapy and present lack of clinically established targeted therapies. Advances in the design of individualized strategies for treatment of TNBC patients require further elucidation, by combined 'omics' approaches, of the molecular mechanisms underlying TNBC phenotypic heterogeneity, and the still poorly understood association of TNBC with BRCA1 mutations. An overview is here presented on TNBC profiling in terms of expression signatures, within the functional genomic breast tumor classification, and ongoing efforts toward identification of new therapy targets and bioimaging markers. Due to the complexity of aberrant molecular patterns involved in expression, pathological progression and biological/clinical heterogeneity, the search for novel TNBC biomarkers and therapy targets requires collection of multi-dimensional data sets, use of robust multivariate data analysis techniques and development of innovative systems biology approaches.

Integrating gene expression and GO classification for PCA by preclustering.

BACKGROUND: Gene expression data can be analyzed by summarizing groups of individual gene expression profiles based on GO annotation information. The mean expression profile per group can then be used to identify interesting GO categories in relation to the experimental settings. However, the expression profiles present in GO classes are often heterogeneous, i.e., there are several different expression profiles within one class. As a result, important experimental findings can be obscured because the summarizing profile does not seem to be of interest. We propose to tackle this problem by finding homogeneous subclasses within GO categories: preclustering. RESULTS: Two microarray datasets are analyzed. First, a selection of genes from a well-known Saccharomyces cerevisiae dataset is used. The GO class "cell wall organization and biogenesis" is shown as a specific example. After preclustering, this term can be associated with different phases in the cell cycle, where it could not be associated with a specific phase previously. Second, a dataset of differentiation of human Mesenchymal Stem Cells (MSC) into osteoblasts is used. For this dataset results are shown in which the GO term "skeletal development" is a specific example of a heterogeneous GO class for which better associations can be made after preclustering. The Intra Cluster Correlation (ICC), a measure of cluster tightness, is applied to identify relevant clusters. CONCLUSIONS: We show that this method leads to an improved interpretability of results in Principal Component Analysis.

Combination of feature-reduced MR spectroscopic and MR imaging data for improved brain tumor classification.

The purpose of this paper is to evaluate the effect of the combination of magnetic resonance spectroscopic imaging (MRSI) data and magnetic resonance imaging (MRI) data on the classification result of four brain tumor classes. Suppressed and unsuppressed short echo time MRSI and MRI were performed on 24 patients with a brain tumor and four volunteers. Four different feature reduction procedures were applied to the MRSI data: simple quantitation, principal component analysis, independent component analysis and LCModel. Water intensities were calculated from the unsuppressed MRSI data. Features were extracted from the MR images which were acquired with four different contrasts to comply with the spatial resolution of the MRSI. Evaluation was performed by investigating different combinations of the MRSI features, the MRI features and the water intensities. For each data set, the isolation in feature space of the tumor classes, healthy brain tissue and cerebrospinal fluid was calculated and visualized. A test set was used to calculate classification results for each data set. Finally, the effect of the selected feature reduction procedures on the MRSI data was investigated to ascertain whether it was more important than the addition of MRI information. Conclusions are that the combination of features from MRSI data and MRI data improves the classification result considerably when compared with features obtained from MRSI data alone. This effect is larger than the effect of specific feature reduction procedures on the MRSI data. The addition of water intensities to the data set also increases the classification result, although not significantly. We show that the combination of data from different MR investigations can be very important for brain tumor classification, particularly if a large number of tumors are to be classified simultaneously.

A chemometric approach for brain tumor classification using magnetic resonance imaging and spectroscopy.

A new classification approach was developed to improve the noninvasive diagnosis of brain tumors. Within this approach, information is extracted from magnetic resonance imaging and spectroscopy data, from which the relative location and distribution of selected tumor classes in feature space can be calculated. This relative location and distribution is used to select the best information extraction procedure, to identify overlapping tumor classes, and to calculate probabilities of class membership. These probabilities are very important, since they provide information about the reliability of classification and might provide information about the heterogeneity of the tissue. Classification boundaries were calculated by setting thresholds for each investigated tumor class, which enabled the classification of new objects. Results on histopathologically determined tumors are excellent, demonstrated by spatial maps showing a high probability for the correctly identified tumor class and, moreover, low probabilities for other tumor classes.