Shinichi Oeda

Knowledge Discovery and Data Mining in Medicine

Medical databases store diagnostic information based on patients’ medical records. Because of deficits in patients’ medical records, medical databases do not provide all the required information for learning algorithms. Moreover, we may meet some contradictory cases, in which the pattern of input signals is the same, but the pattern of output signals is different. Learning algorithms cannot correctly classify such cases. Even medical doctors require more information to make the final diagnosis. In this chapter, we describe three methods of classifying medical databases based on neural networks and genetic programming (GP). To verify the effectiveness of our proposed methods, we apply them to real medical databases and prove their high classification capability. We also introduce techniques for extracting If-Then rules from the trained networks.

A learning method of immune multi-agent neural networks

Many different learning algorithms for neural networks have been developed, with advantages offered in terms of network structure, initial values of some parameters, learning speed, and learning accuracy. If we train the networks only on good examples, without noise and shortage, the neural network can be trained to classify, with reasonable accuracy, target patterns or random patterns, but not both. To solve this problem, we propose a learning method of immune multi-agent neural networks (IMANNs), which have agents of macrophages, B-cells and T-cells. Each agent employs a different type of neural network. Because the agents work cooperatively and competitively, IMANNs can automatically classify the training dataset into some subclasses. In this paper, two types of IMANNs are described and their classification capabilities are compared. In order to verify the effectiveness of our proposed method, we used two datasets: the dataset of the MONK’s problem (as a traditional classification problem) and a dataset from a medical diagnosis problem (hepatobiliary disorders).

Time series data classification using recurrent neural network with ensemble learning

In statistics and signal processing, a time series is a sequence of data points, measured typically at successive times, spaced apart at uniform time intervals. Time series prediction is the use of a model to predict future events based on known past events; to predict future data points before they are measured. Solutions in such cases can be provided by non-parametric regression methods, of which each neural network based predictor is a class. As a learning method of time series data with neural network, Elman type Recurrent Neural Network has been known. In this paper, we propose the multi RNN. In order to verify the effectiveness of our proposed method, we experimented by the simple artificial data and the heart pulse wave data.

Immune Multi Agent Neural Network and Its Application to the Coronary Heart Disease Database

Medical databases hold diagnostic information based on patient medical records. However, these medical records may not always hold enough information for standard learning algorithms. Sometimes contradictory cases may occur, in which the pattern of input signals is the same, but the pattern of output signals is starkly different. In this paper, we apply a learning method of the immune multi agent neural networks (IMANNs) to the medical disease databases. IMANNs have agents of the macrophages, B-cells, and T-cells. Each agent employs a different type of neural networks. Because their agents work cooperatively and competitively, IMANNs can classify training dataset into some subsets automatically, and successively each B-cell agent trains specially for the divided training subset. In order to verify the effectiveness of our proposed method, we tested the coronary heart disease database as medical databases.

A Classification Method of Medical Database by Immune Multi-agent Neural Networks with Planar Lattice Architecture

A classification capability of multi expert neural networks shows some outstanding results in training a set of real training cases. Multi expert neural networks work to operate with each other based on the concept of multi agents. However, we may meet a problem of an optimal number of agents. In this paper, we apply the Planar Lattice Neural Network to find an optimal number while classifying a set of training cases. To verify the effectiveness, we report the experimental results for hepatobiliary disorders medical databases.

Neural Networks Applied to Medical Data for Prediction of Patient Outcome

Prediction is vital in clinical fields, because it influences decision making for treatment and resource allocation. At present, medical records are readily accessible from hospital information systems. Based on the analysis of medical records, a number of predictive models have been developed to support the prediction of patient outcome. However, predictive models that achieve the desired predictive performance are few and far between. Approaches to developing predictive models vary from traditional statistical methods to artificial intelligence methods. Multivariate regression models, particularly logistic regression, are the most commonly applied models, and have been for some time. As a potential alternative to multivariate regression models, interest in the use of neural networks (NNs) has recently been expressed [1, 9, 11, 14]. Because each modeling method has its own strengths and limitations [2, 8, 9, 11, 14], it is hard to determine which modeling method is most suitable for the prediction of patient outcome. Medical data are known to have their own unique characteristics, which may impede the development of a good predictive model [7]. Comparative studies using real medical data are expected to pave the way for more effective modeling methods. In this chapter, we describe the capability of NNs applied to medical data for the prediction of patient outcome. Firstly, we applied a simple three-layer backpropagation NN to a dataset of intensive care unit (ICU) patients [12, 13] to develop a predictive model that estimates the probability of nosocomial infection. The predictive performance of the NN was compared with that of logistic regression using the cross-validation method. Secondly, we invented a method of modeling time sequence data for prediction using multiple NNs. Based on the dataset of ICU patients, we examined whether multiple NNs outperform both logistic regression and the application of a single NN in the long-term prediction of nosocomial infection. According to the results of these studies, careful preparation of datasets improves the predictive performance of

A Classification Capability of Reflective Neural Networks in Medical Databases

Medical database recorded diagnostic information based on a patient’s medical card. However, each card does not fulfill all the required information for learning algorithm. Reflective Neural Network has an outstanding classification capability, even if such records with shortage exist in a set of training cases. Reflective Neural Network is based on the network module concept. There are two kinds of network modules; an allocation module to distribute a training case and some classification modules to classify a subset of training cases. Each classification module consists of a monitor neural network and a worker neural network. The monitor neural network estimates how conformable the worker neural network is to a given training case. Moreover, the training case is distributed over different classification modules. These classification modules compete with each other in the classification task. In this paper, we report the classification capability in a medical database on the patients in ICUs.

A Proposal of Immune Multi-agent Neural Networks and Its Application to Medical Diagnostic System for Hepatobiliary Disorders

Both immune system and neural network are complex biological systems. These systems are capable of learning, memory, and pattern recognition. Many classification algorithms have been developed in a field of the information processing. In this paper, we propose the immune multi agent neural networks where each immune agent employs different neural networks to handle a subset of training cases. This proposed method is limited to the behaviors of the macrophage, B-cell, and T-cell to realize a good classification capability. To verify the validity and effectiveness of the proposed method, we developed a diagnostic system for hepatobiliary disorders.

An adaptive evolutional neuro learning method using genetic search and extraction of rules from trained networks

BP learning is widely known to perform good classification for given training data. However, there is a kind of noise or inconsistent knowledge in training cases. In this case, a neural network will not converge. To avoid such a problem, we propose an adaptive evolutional neuro learning method to handle a subset of the complete set of training cases. This method has a sufficient adaptive ability like a living things evolutionary process based on Darwinian Genetic Inheritance. In this method, the network structure is determined by genetic search for each generation and the connection weights and learning parameters determined by BP learning are not inherited. Furthermore, we tried to extract rules from the trained network. To verify the validity and effectiveness of the proposed method, we develop the diagnostic system for hepatobiliary disorders.

Construction of emotional space from facial expression by parallel sand glass type neural networks

We proposed a parallel sand glass type neural network. The network model can learn some peoples facial expressions. This network has some five-layered neural networks and hidden neurons in each third layer are connected. Our proposed network model can construct an emotional space based on facial expressions. This emotional space is similar to the emotional circle of Ekman et al. (1972). However, there is a weak point in learning facial. expressions of a lot of people, because the learning algorithm requires a five-layered network for learning faces of a person. We improved the learning algorithm in only a two-combined network. To verify the validity of proposed network model, we experimented facial expressions of some people.