Supported Diagnosis of Attention Deficit and Hyperactivity Disorder from EEG Based on Interpretable Kernels for Hidden Markov Models.

As a neurodevelopmental pathology, Attention Deficit Hyperactivity Disorder (ADHD) mainly arises during childhood. Persistent patterns of generalized inattention, impulsivity, or hyperactivity characterize ADHD that may persist into adulthood. The conventional diagnosis relies on clinical observational processes yielding high rates of overdiagnosis due to varying interpretations among specialists or missing information. Although several studies have designed objective behavioral features to overcome such an issue, they lack significance. Despite electroencephalography (EEG) analyses extracting alternative biomarkers using signal processing techniques, the nonlinearity and nonstationarity of EEG signals restrain performance and generalization of hand-crafted features. This work proposes a methodology to support ADHD diagnosis by characterizing EEG signals from hidden Markov models (HMM), classifying subjects based on similarity measures for probability functions, and spatially interpreting the results using graphic embeddings of stochastic dynamic models. The methodology learns a single HMM for EEG signal from each patient, so favoring the inter-subject variability. Then, the Probability Product Kernel, specifically developed for assessing the similarity between HMMs, fed a support vector machine that classifies subjects according to their stochastic dynamics. Lastly, the kernel variant of Principal Component Analysis provided a means to visualize the EEG transitions in a two-dimensional space, evidencing dynamic differences between ADHD and Healthy Control children. From the electrophysiological perspective, we recorded EEG under the Stop Signal Task modified with reward levels, which considers cognitive features of interest as insufficient motivational circuits recruitment. The methodology compares the supported diagnosis in two EEG channel setups (whole channel set and channels of interest in frontocentral area) and four frequency bands (Theta, Alpha, Beta rhythms, and a wideband). Results evidence an accuracy rate of 97.0% in the Beta band and in the channels where previous works found error-related negativity events. Such accuracy rate strongly supports the dual pathway hypothesis and motivational deficit concerning the pathophysiology of ADHD. It also demonstrates the utility of joining inhibitory and motivational paradigms with dynamic EEG analysis into a noninvasive and affordable diagnostic tool for ADHD patients.

Instance-Based Representation Using Multiple Kernel Learning for Predicting Conversion to Alzheimer Disease.

The early detection of Alzheimer's disease and quantification of its progression poses multiple difficulties for machine learning algorithms. Two of the most relevant issues are related to missing data and results interpretability. To deal with both issues, we introduce a methodology to predict conversion of mild cognitive impairment patients to Alzheimer's from structural brain MRI volumes. First, we use morphological measures of each brain structure to build an instance-based feature mapping that copes with missed follow-up visits. Then, the extracted multiple feature mappings are combined into a single representation through the convex combination of reproducing kernels. The weighting parameters per structure are tuned based on the maximization of the centered-kernel alignment criterion. We evaluate the proposed methodology on a couple of well-known classification machines employing the ADNI database devoted to assessing the combined prognostic value of several AD biomarkers. The obtained experimental results show that our proposed method of Instance-based representation using multiple kernel learning enables detecting mild cognitive impairment as well as predicting conversion to Alzheimers disease within three years from the initial screening. Besides, the brain structures with larger combination weights are directly related to memory and cognitive functions.

Kernel-based atlas image selection for brain tissue segmentation.

We propose a new Kernel-based Atlas Image Selection computed in the Embedding Representation space (termed KAISER) aiming to support labeling of brain tissue on 3D magnetic resonance (MR) images. KAISER approach provides efficient feature extraction from MR volumes based on an introduced inter-slice kernel (ISK). Thus, using the ISK matrix eigendecomposition, the inherent structure of data distribution is accentuated through estimation of low dimensional compact space where every pair-wise image similarity can be better measured. We compare our proposal against the whole-population atlas, randomly and demographically selected multiatlas approaches in a four-tissue image labeling task. Obtained results show that the KAISER approach outperforms other alternative techniques (98% Dice index similarity against 94%), while exhibiting better repeatability.

Tensor-product kernel-based representation encoding joint MRI view similarity.

To support 3D magnetic resonance image (MRI) analysis, a marginal image similarity (MIS) matrix holding MR inter-slice relationship along every axis view (Axial, Coronal, and Sagittal) can be estimated. However, mutual inference from MIS view information poses a difficult task since relationships between axes are nonlinear. To overcome this issue, we introduce a Tensor-Product Kernel-based Representation (TKR) that allows encoding brain structure patterns due to patient differences, gathering all MIS matrices into a single joint image similarity framework. The TKR training strategy is carried out into a low dimensional projected space to get less influence of voxel-derived noise. Obtained results for classifying the considered patient categories (gender and age) on real MRI database shows that the proposed TKR training approach outperforms the conventional voxel-wise sum of squared differences. The proposed approach may be useful to support MRI clustering and similarity inference tasks, which are required on template-based image segmentation and atlas construction.

Influence of anisotropic white matter modeling on EEG source localization.

We study the influence of the anisotropic white matter within the ElectroEncephaloGraphy source localization problem. To this end, we consider three cases of the anisotropic white matter modeled in two concrete cases: by fixed or variable ratio. We extract information about highly anisotropic areas of the white matter from real Diffusion Weighted Imaging data. To validate the compared anisotropic models, we introduce the localization dipole and orientation errors. Obtained results show that the white matter model with a fixed anisotropic ratio leads to values of dipole localization error close to 1cm and may be enough in those cases avoiding localized analysis of neural brain activity. In contrast, modeling based on the anisotropic variable rate assumption becomes important in tasks regarding analysis and localization of deep sources neighboring the white matter tissue.

Local binary fitting energy solution by graph cuts for MRI segmentation.

This paper proposes a new solution for local binary fitting energy minimization based on graph cuts for automatic brain structure segmentation on magnetic resonance images. The approach establishes an effective way to embed the energy formulation into a directed graph, such that the energy is minimized by maximizing the graph flow. Proposed and conventional solutions are compared by segmenting the well-known BrainWeb synthetic brain Magnetic Resonance Imaging database. Achieved results show an improvement on the computational cost (about 10 times shorter) while maintaining the segmentation accuracy (96%).

Extraction of stationary components in biosignal discrimination.

Biosignal recordings are widely used in the medical environment to support the evaluation and the diagnosis of pathologies. Nevertheless, the main difficulty lies in the non-stationary behavior of the biosignals, difficulting the obtention of patterns characterizing the changes in physiological or pathological states. Thus, the obtention of the stationary and non-stationary components of a biosignal is still an open issue. This work proposes a methodology to detect time-homogeneities based on time-frequency analysis with aim to extract the non-stationary behavior of the biosignal. Results show an increase in the stationarity and in the distance between classes of the reconstructions from the enhanced time-frequency representations. The stationary components extracted with the proposed approach can be used to solve biosignal classification problems.