The analysis of functional magnetic resonance imaging (fMRI) data is complicated by the presence of a mixture of many sources of signal and noise. Independent component analysis (ICA) can separate these mixtures into independent components, each of which contains maximal information from a single, independent source of signal, whether from noise or from a discrete physiological or neural system. ICA typically generates a large number of components for each subject imaged, however, and therefore it generates a vast number of components across all of the subjects imaged in an fMRI dataset. The practical implementation of ICA has been limited by the difficulty in discerning which of these many components are spurious and which are reproducible, either within or across individuals of the dataset. We have developed a novel clustering algorithm, termed “Partner‐Matching” (PM), which identifies automatically the independent components that are reproducible either within or between subjects. It identifies those components by clustering them according to robust measures of similarity in their spatial configurations either across different subjects of an fMRI dataset, within a single subject scanned across multiple scanning sessions, or within an individual subject scanned across multiple runs within a single scanning session. We demonstrate the face validity of our algorithm by applying it to the analysis of three fMRI datasets acquired in 13 healthy adults performing simple auditory, motor, and visual tasks. From among 50 independent components generated for each subject, our PM algorithm automatically identified, across all 13 subjects, components representing activity within auditory, motor, and visual cortices, respectively, as well as numerous other reproducible components outside of primary sensory and motor cortices, in functionally connected circuits that subserve higher‐order cognitive functions, even in these simple tasks. Hum Brain Mapp, 2008.
7 February 2008 in IEEE Transactions on Neural Networks
It is well known that least absolute deviation (LAD) criterion or -norm used for estimation of parameters is characterized by robustness, i.e., the estimated parameters are totally resistant (insensitive) to large changes in the sampled data. This is an extremely useful feature, especially, when the sampled data are known to be contaminated by occasionally occurring outliers or by spiky noise. In our previous works, we have proposed the least absolute deviation neural network (LADNN) to solve unconstrained LAD problems. The theoretical proofs and numerical simulations have shown that the LADNN is Lyapunov-stable and it can globally converge to the exact solution to a given unconstrained LAD problem. We have also demonstrated its excellent application value in time-delay estimation. More generally, a practical LAD application problem may contain some linear constraints, such as a set of equalities and/or inequalities, which is called constrained LAD problem, whereas the unconstrained LAD can be considered as a special form of the constrained LAD. In this paper, we present a new neural network called constrained least absolute deviation neural network (CLADNN) to solve general constrained LAD problems. Theoretical proofs and numerical simulations demonstrate that the proposed CLADNN is Lyapunov stable and globally converges to the exact solution to a given constrained LAD problem, independent of initial values. The numerical simulations have also illustrated that the proposed CLADNN can be used to robustly estimate parameters for nonlinear curve fitting, which is extensively used in signal and image processing.
19 July 2004 in IEEE Transactions on Biomedical Engineering
Nonlinear feedback induces oscillation, whereas dynamic equilibrium between positive and negative nonlinear feedback generates rhythm. Physiological rhythms are central to life. No absolutely stable or periodic rhythm exists in living tissues. It has been extensively reported that many rhythms in human and animal organs, such as the heart and brain, are, in fact chaotic. The aim of this paper is to investigate whether the migrating myoelectrical complex (MMC) of the stomach was chaotic. The study was performed in eight healthy female hound dogs (15-22 kg), implanted with four pairs of bipolar electrodes on the serosa of the stomach along the greater curvature. After the dogs were completely recovered from the surgery, one complete cycle of gastric MMC was recorded from the serosal electrodes. Using Takens' embedding theorem, two parameters reflecting chaotic behavior, the attractor and the Lyapunov exponent of the myoelectrical recording, were reconstructed and computed, respectively. Statistical analysis was performed to investigate the difference in the Lyapunov exponents among different phases of the MMC. The results show that the MMC of the stomach is chaotic. Different phases of the MMC are characterized with different shapes of the attractors and different values of Lyapunov exponents. The characteristic chaotic behavior of the gastric MMC may be utilized for the identification of different phases.
For attention deficit hyperactivity disorder (ADHD), a common neurological disease, accurate identification is the basis for treatment. In this paper, a novel end-to-end representation learning framework for ADHD classification of functional magnetic resonance imaging (fMRI) sequences is proposed. With such a framework, the complexity of the sequence representation learning neural network decreases, the overfitting problem of deep learning for small samples cases is solved effectively, and superior classification performance is achieved. Specifically, a data conversion module was designed to convert a two-dimensional sequence into a three-dimensional image, which expands the modeling area and greatly reduces the computational complexity. The transfer learning method was utilized to freeze or fine-tune the parameters of the pre-trained neural network to reduce the risk of overfitting in the cases with small samples. Hierarchical feature extraction can be performed automatically by combining the sequence representation learning modules with a weighted cross-entropy loss. Experiments were conducted both with individual imaging sites and combining them, and the results showed that the classification average accuracies with the proposed framework were 73.73% and 72.02%, respectively, which are much higher than those of the existing methods.