Highly scalable and robust rule learner: performance evaluation and comparison.

Business intelligence and bioinformatics applications increasingly require the mining of datasets consisting of millions of data points, or crafting real-time enterprise-level decision support systems for large corporations and drug companies. In all cases, there needs to be an underlying data mining system, and this mining system must be highly scalable. To this end, we describe a new rule learner called DataSqueezer. The learner belongs to the family of inductive supervised rule extraction algorithms. DataSqueezer is a simple, greedy, rule builder that generates a set of production rules from labeled input data. In spite of its relative simplicity, DataSqueezer is a very effective learner. The rules generated by the algorithm are compact, comprehensible, and have accuracy comparable to rules generated by other state-of-the-art rule extraction algorithms. The main advantages of DataSqueezer are very high efficiency, and missing data resistance. DataSqueezer exhibits log-linear asymptotic complexity with the number of training examples, and it is faster than other state-of-the-art rule learners. The learner is also robust to large quantities of missing data, as verified by extensive experimental comparison with the other learners. DataSqueezer is thus well suited to modern data mining and business intelligence tasks, which commonly involve huge datasets with a large fraction of missing data.     

A new synaptic plasticity rule for networks of spiking neurons

In this paper, we describe a new Synaptic Plasticity Activity Rule (SAPR) developed for use in networks of spiking neurons. Such networks can be used for simulations of physiological experiments as well as for other computations like image analysis. Most synaptic plasticity rules use artificially defined functions to modify synaptic connection strengths. In contrast, our rule makes use of the existing postsynaptic potential values to compute the value of adjustment. The network of spiking neurons we consider consists of excitatory and inhibitory neurons. Each neuron is implemented as an integrate-and-fire model that accurately mimics the behavior of biological neurons. To test performance of our new plasticity rule we designed a model of a biologically-inspired signal processing system, and used it for object detection in eye images of diabetic retinopathy patients, and lung images of cystic fibrosis patients. The results show that the network detects the edges of objects within an image, essentially segmenting it. Our ultimate goal, however, is not the development of an image segmentation tool that would be more efficient than nonbiological algorithms, but developing a physiologically correct neural network model that could be applied to a wide range of neurological experiments. We decided to validate the SAPR by using it in a network of spiking neurons for image segmentation because it is easy to visually assess the results. An important thing is that image segmentation is done in an entirely unsupervised way.     

Epileptic seizure detection.

In this study, radial basis function (RBF) neural networks are used to identify seizure or preseizure states. As input to the RBF networks the study used raw EEG data, coefficients from a Fourier transform, and wavelet decomposition of the raw data. An RBF network consists of an input layer, a single hidden layer, and an output node. The use of half-second windows of raw data as input demonstrates the ability of the RBF network to learn differences in the patterns of ictal and interictal EEG data without feature extraction. Wavelet decomposition of the narrow window of raw data improves performance while transformation of a wider window, up to about five seconds, improves it even further. The ability of wavelet decomposition to transform five seconds of raw data into a vector of manageable length without substantial loss of relevant information makes it an effective tool for preprocessing EEG data     

A knowledge discovery approach to diagnosing myocardial perfusion

Discusses applying a six-step discovery process to a database of SPECT bull's-eye maps of the heart. Visual assessment of clinical diagnostic images is observer-dependent. Thus, much effort is expended to computerize the process of diagnosis so it is less dependent on the observer, especially when the observer is not experienced. A large number of images to be evaluated (as in SPECT myocardial perfusion studies: approximately 15 oblique “slices,” 15 oblique/sagittal, and 15 oblique/coronal, both in stress and rest, which comes to nearly 100 2-D images per patient) forced the creation of more “comprehensive” images; namely, the bull's-eye perfusion maps. Using these maps, the authors showed that it is possible to differentiate the patients with coronary artery disease (one- or two-vessel) from the patients with low probability of the disease (normals). In the future, features other than those used in this work will be used; for instance, a feature representing the area of “abnormal” myocardium, available in most previously mentioned algorithms for “normative” evaluation of bull's-eye maps. In the course of this work, the authors also came up with methods that can accurately extract the ROIs from an image where a thresholding method cannot be used     

Identifying coronary stenosis using an image-recognition neuralnetwork

The present study represents an attempt to improve the separation of specific high-risk coronary stenosis from lower-risk conditions by employing an image-recognition neural network. The system mimics the visual reading of scintigraphs from raw digitized data, with the added benefit of a computerized classification system. It models the human retina as the sensing organ that processes the image signals and forwards them to the brain, where the outputs from each visual segment are processed to produce a recognition code. In the application described here, the recognition code classifies a scintigraphic image as demonstrating normal myocardial perfusion, or a perfusion pattern consistent with single-vessel, multiple-vessel, or left-anterior descending coronary artery stenosis. The input images are from clinically performed postexercise planar myocardial perfusion scintigraphs as produced in many clinical laboratories     

Use of neural networks in detecting cardiac diseases fromechocardiographic images

The usefulness of backpropagation neural networks in the analysis of two-dimensional echocardiographic (2DE) images has been evaluated. The gray-scale levels from 2DE images directly correspond to the intensity of echo signal from cardiac tissue, providing visual texture and allowing qualitative and quantitative analysis of myocardial tissue. A subject population consisting of 11 normal, 7 hypertrophic cardiomyopathy, and 11 myocardial infarction patients was studied. Two types of backpropagational neural networks were used: fully connected, and patterned. In the fully connected network, the outputs of neurodes in each layer are connected to the inputs of all neurodes in the following layer. In the patterned network, only neurodes within a defined neighborhood are connected. The results suggest that the fully connected network provides better classifying performance than the patterned network.     

Synthesis of feedforward networks in supremum error bound

The main result of this paper is a constructive proof of a formula for the upper bound of the approximation error in L/sub /spl infin// (supremum norm) of multidimensional functions by feedforward networks with one hidden layer of sigmoidal units and a linear output. This result is applied to formulate a new method of neural-network synthesis. The result can also be used to estimate complexity of the maximum-error network and/or to initialize that network's weights. An example of the network synthesis is given.     

Strain-induced martensite reversion in 18Cr–8Ni steel – transmission Kikuchi diffraction study

ABSTRACT

Reverse transformation of strain-induced martensite (SIM) was studied in the 18Cr–8Ni stainless steel. Microstructure analysis was performed on samples in an as-deformed state and after reversion annealing at 873 and 973?K, using the transmission Kikuchi diffraction. The primary, as well as reversed, austenite possesses the Kurdjumov–Sachs crystallographic orientation relationship, with respect to the SIM. The reverted austenite keeps one orientation within all reverted grains, regardless of the applied heating procedure, where the amount of reverted austenite depends only on the annealing temperature.