Attributed Multi-Order Graph Convolutional Network for Heterogeneous Graphs.

Heterogeneous graph neural networks play a crucial role in discovering discriminative node embeddings and relations from multi-relational networks. One of the key challenges in heterogeneous graph learning lies in designing learnable meta-paths, which significantly impact the quality of learned embeddings. In this paper, we propose an Attributed Multi-Order Graph Convolutional Network (AMOGCN), which automatically explores meta-paths that involve multi-hop neighbors by aggregating multi-order adjacency matrices. The proposed model first constructs different orders of adjacency matrices from manually designed node connections. Next, AMOGCN fuses these various orders of adjacency matrices to create an intact multi-order adjacency matrix. This process is supervised by the node semantic information, which is extracted from the node homophily evaluated by attributes. Eventually, we employ a one-layer simplifying graph convolutional network with the learned multi-order adjacency matrix, which is equivalent to the cross-hop node information propagation with multi-layer graph neural networks. Substantial experiments reveal that AMOGCN achieves superior semi-supervised classification performance compared with state-of-the-art competitors.

AGNN: Alternating Graph-Regularized Neural Networks to Alleviate Over-Smoothing.

Graph convolutional network (GCN) with the powerful capacity to explore graph-structural data has gained noticeable success in recent years. Nonetheless, most of the existing GCN-based models suffer from the notorious over-smoothing issue, owing to which shallow networks are extensively adopted. This may be problematic for complex graph datasets because a deeper GCN should be beneficial to propagating information across remote neighbors. Recent works have devoted effort to addressing over-smoothing problems, including establishing residual connection structure or fusing predictions from multilayer models. Because of the indistinguishable embeddings from deep layers, it is reasonable to generate more reliable predictions before conducting the combination of outputs from various layers. In light of this, we propose an alternating graph-regularized neural network (AGNN) composed of graph convolutional layer (GCL) and graph embedding layer (GEL). GEL is derived from the graph-regularized optimization containing Laplacian embedding term, which can alleviate the over-smoothing problem by periodic projection from the low-order feature space onto the high-order space. With more distinguishable features of distinct layers, an improved Adaboost strategy is utilized to aggregate outputs from each layer, which explores integrated embeddings of multi-hop neighbors. The proposed model is evaluated via a large number of experiments including performance comparison with some multilayer or multi-order graph neural networks, which reveals the superior performance improvement of AGNN compared with the state-of-the-art models.

Quantification of Kuramoto Coupling Between Intrinsic Brain Networks Applied to fMRI Data in Major Depressive Disorder.

Organized patterns of system-wide neural activity adapt fluently within the brain to adjust behavioral performance to environmental demands. In major depressive disorder (MD), markedly different co-activation patterns across the brain emerge from a rather similar structural substrate. Despite the application of advanced methods to describe the functional architecture, e.g., between intrinsic brain networks (IBNs), the underlying mechanisms mediating these differences remain elusive. Here we propose a novel complementary approach for quantifying the functional relations between IBNs based on the Kuramoto model. We directly estimate the Kuramoto coupling parameters (K) from IBN time courses derived from empirical fMRI data in 24 MD patients and 24 healthy controls. We find a large pattern with a significant number of Ks depending on the disease severity score Hamilton D, as assessed by permutation testing. We successfully reproduced the dependency in an independent test data set of 44 MD patients and 37 healthy controls. Comparing the results to functional connectivity from partial correlations (FC), to phase synchrony (PS) as well as to first order auto-regressive measures (AR) between the same IBNs did not show similar correlations. In subsequent validation experiments with artificial data we find that a ground truth of parametric dependencies on artificial regressors can be recovered. The results indicate that the calculation of Ks can be a useful addition to standard methods of quantifying the brain's functional architecture.

Heterogeneous Graphical Granger Causality by Minimum Message Length.

The heterogeneous graphical Granger model (HGGM) for causal inference among processes with distributions from an exponential family is efficient in scenarios when the number of time observations is much greater than the number of time series, normally by several orders of magnitude. However, in the case of "short" time series, the inference in HGGM often suffers from overestimation. To remedy this, we use the minimum message length principle (MML) to determinate the causal connections in the HGGM. The minimum message length as a Bayesian information-theoretic method for statistical model selection applies Occam's razor in the following way: even when models are equal in their measure of fit-accuracy to the observed data, the one generating the most concise explanation of data is more likely to be correct. Based on the dispersion coefficient of the target time series and on the initial maximum likelihood estimates of the regression coefficients, we propose a minimum message length criterion to select the subset of causally connected time series with each target time series and derive its form for various exponential distributions. We propose two algorithms-the genetic-type algorithm (HMMLGA) and exHMML to find the subset. We demonstrated the superiority of both algorithms in synthetic experiments with respect to the comparison methods Lingam, HGGM and statistical framework Granger causality (SFGC). In the real data experiments, we used the methods to discriminate between pregnancy and labor phase using electrohysterogram data of Islandic mothers from Physionet databasis. We further analysed the Austrian climatological time measurements and their temporal interactions in rain and sunny days scenarios. In both experiments, the results of HMMLGA had the most realistic interpretation with respect to the comparison methods. We provide our code in Matlab. To our best knowledge, this is the first work using the MML principle for causal inference in HGGM.

Intrinsic brain activity of cognitively normal older persons resembles more that of patients both with and at risk for Alzheimer's disease than that of healthy younger persons.

In Alzheimer's disease (AD), recent findings suggest that amyloid-ß (Aß)-pathology might start 20-30 years before first cognitive symptoms arise. To account for age as most relevant risk factor for sporadic AD, it has been hypothesized that lifespan intrinsic (i.e., ongoing) activity of hetero-modal brain areas with highest levels of functional connectivity triggers Aß-pathology. This model induces the simple question whether in older persons without any cognitive symptoms intrinsic activity of hetero-modal areas is more similar to that of symptomatic patients with AD or to that of younger healthy persons. We hypothesize that due to advanced age and therefore potential impact of pre-clinical AD, intrinsic activity of older persons resembles more that of patients than that of younger controls. We tested this hypothesis in younger (ca. 25 years) and older healthy persons (ca. 70 years) and patients with mild cognitive impairment and AD-dementia (ca. 70 years) by the use of resting-state functional magnetic resonance imaging, distinct measures of intrinsic brain activity, and different hierarchical clustering approaches. Independently of applied methods and involved areas, healthy older persons' intrinsic brain activity was consistently more alike that of patients than that of younger controls. Our result provides evidence for larger similarity in intrinsic brain activity between healthy older persons and patients with or at-risk for AD than between older and younger ones, suggesting a significant proportion of pre-clinical AD cases in the group of cognitively normal older people. The observed link of aging and AD with intrinsic brain activity supports the view that lifespan intrinsic activity may contribute critically to the pathogenesis of AD.

Robust automated detection of microstructural white matter degeneration in Alzheimer's disease using machine learning classification of multicenter DTI data.

Diffusion tensor imaging (DTI) based assessment of white matter fiber tract integrity can support the diagnosis of Alzheimer's disease (AD). The use of DTI as a biomarker, however, depends on its applicability in a multicenter setting accounting for effects of different MRI scanners. We applied multivariate machine learning (ML) to a large multicenter sample from the recently created framework of the European DTI study on Dementia (EDSD). We hypothesized that ML approaches may amend effects of multicenter acquisition. We included a sample of 137 patients with clinically probable AD (MMSE 20.6±5.3) and 143 healthy elderly controls, scanned in nine different scanners. For diagnostic classification we used the DTI indices fractional anisotropy (FA) and mean diffusivity (MD) and, for comparison, gray matter and white matter density maps from anatomical MRI. Data were classified using a Support Vector Machine (SVM) and a Naïve Bayes (NB) classifier. We used two cross-validation approaches, (i) test and training samples randomly drawn from the entire data set (pooled cross-validation) and (ii) data from each scanner as test set, and the data from the remaining scanners as training set (scanner-specific cross-validation). In the pooled cross-validation, SVM achieved an accuracy of 80% for FA and 83% for MD. Accuracies for NB were significantly lower, ranging between 68% and 75%. Removing variance components arising from scanners using principal component analysis did not significantly change the classification results for both classifiers. For the scanner-specific cross-validation, the classification accuracy was reduced for both SVM and NB. After mean correction, classification accuracy reached a level comparable to the results obtained from the pooled cross-validation. Our findings support the notion that machine learning classification allows robust classification of DTI data sets arising from multiple scanners, even if a new data set comes from a scanner that was not part of the training sample.

Prediction of Alzheimer's disease using individual structural connectivity networks.

Alzheimer's disease (AD) progressively degrades the brain's gray and white matter. Changes in white matter reflect changes in the brain's structural connectivity pattern. Here, we established individual structural connectivity networks (ISCNs) to distinguish predementia and dementia AD from healthy aging in individual scans. Diffusion tractography was used to construct ISCNs with a fully automated procedure for 21 healthy control subjects (HC), 23 patients with mild cognitive impairment and conversion to AD dementia within 3 years (AD-MCI), and 17 patients with mild AD dementia. Three typical pattern classifiers were used for AD prediction. Patients with AD and AD-MCI were separated from HC with accuracies greater than 95% and 90%, respectively, irrespective of prediction approach and specific fiber properties. Most informative connections involved medial prefrontal, posterior parietal, and insular cortex. Patients with mild AD were separated from those with AD-MCI with an accuracy of approximately 85%. Our finding provides evidence that ISCNs are sensitive to the impact of earliest stages of AD. ISCNs may be useful as a white matter-based imaging biomarker to distinguish healthy aging from AD.

Decoding an individual's sensitivity to pain from the multivariate analysis of EEG data.

The perception of pain is characterized by its tremendous intra- and interindividual variability. Different individuals perceive the very same painful event largely differently. Here, we aimed to predict the individual pain sensitivity from brain activity. We repeatedly applied identical painful stimuli to healthy human subjects and recorded brain activity by using electroencephalography (EEG). We applied a multivariate pattern analysis to the time-frequency transformed single-trial EEG responses. Our results show that a classifier trained on a group of healthy individuals can predict another individual's pain sensitivity with an accuracy of 83%. Classification accuracy depended on pain-evoked responses at about 8 Hz and pain-induced gamma oscillations at about 80 Hz. These results reveal that the temporal-spectral pattern of pain-related neuronal responses provides valuable information about the perception of pain. Beyond, our approach may help to establish an objective neuronal marker of pain sensitivity which can potentially be recorded from a single EEG electrode.

Feature selection methods for characterizing and classifying adaptive Sustainable Flood Retention Basins.

The European Union's Flood Directive 2007/60/EC requires member states to produce flood risk maps for all river basins and coastal areas at risk of flooding by 2013. As a result, flood risk assessments have become an urgent challenge requiring a range of rapid and effective tools and approaches. The Sustainable Flood Retention Basin (SFRB) concept has evolved to provide a rapid assessment technique for impoundments, which have a pre-defined or potential role in flood defense and diffuse pollution control. A previous version of the SFRB survey method developed by the co-author Scholz in 2006 recommends gathering of over 40 variables to characterize an SFRB. Collecting all these variables is relatively time-consuming and more importantly, these variables are often correlated with each other. Therefore, the objective is to explore the correlation among these variables and find the most important variables to represent an SFRB. Three feature selection techniques (Information Gain, Mutual Information and Relief) were applied on the SFRB data set to identify the importance of the variables in terms of classification accuracy. Four benchmark classifiers (Support Vector Machine, K-Nearest Neighbours, C4.5 Decision Tree and Naïve Bayes) were subsequently used to verify the effectiveness of the classification with the selected variables and automatically identify the optimal number of variables. Experimental results indicate that our proposed approach provides a simple, rapid and effective framework for variable selection and SFRB classification. Only nine important variables are sufficient to accurately classify SFRB. Finally, six typical cases were studied to verify the performance of the identified nine variables on different SFRB types. The findings provide a rapid scientific tool for SFRB assessment in practice. Moreover, the generic value of this tool allows also for its wide application in other areas.

Automated detection of brain atrophy patterns based on MRI for the prediction of Alzheimer's disease.

Subjects with mild cognitive impairment (MCI) have an increased risk to develop Alzheimer's disease (AD). Voxel-based MRI studies have demonstrated that widely distributed cortical and subcortical brain areas show atrophic changes in MCI, preceding the onset of AD-type dementia. Here we developed a novel data mining framework in combination with three different classifiers including support vector machine (SVM), Bayes statistics, and voting feature intervals (VFI) to derive a quantitative index of pattern matching for the prediction of the conversion from MCI to AD. MRI was collected in 32 AD patients, 24 MCI subjects and 18 healthy controls (HC). Nine out of 24 MCI subjects converted to AD after an average follow-up interval of 2.5 years. Using feature selection algorithms, brain regions showing the highest accuracy for the discrimination between AD and HC were identified, reaching a classification accuracy of up to 92%. The extracted AD clusters were used as a search region to extract those brain areas that are predictive of conversion to AD within MCI subjects. The most predictive brain areas included the anterior cingulate gyrus and orbitofrontal cortex. The best prediction accuracy, which was cross-validated via train-and-test, was 75% for the prediction of the conversion from MCI to AD. The present results suggest that novel multivariate methods of pattern matching reach a clinically relevant accuracy for the a priori prediction of the progression from MCI to AD.

Enhancing instance-based classification with local density: a new algorithm for classifying unbalanced biomedical data.

MOTIVATION: Classification is an important data mining task in biomedicine. In particular, classification on biomedical data often claims the separation of pathological and healthy samples with highest discriminatory performance for diagnostic issues. Even more important than the overall accuracy is the balance of a classifier, particularly if datasets of unbalanced class size are examined. RESULTS: We present a novel instance-based classification technique which takes both information of different local density of data objects and local cluster structures into account. Our method, which adopts the basic ideas of density-based outlier detection, determines the local point density in the neighborhood of an object to be classified and of all clusters in the corresponding region. A data object is assigned to that class where it fits best into the local cluster structure. The experimental evaluation on biomedical data demonstrates that our approach outperforms most popular classification methods. AVAILABILITY: The algorithm LCF is available for testing under http://biomed.umit.at/upload/lcfx.zip.