Vitaly Schetinin

A Learning Algorithm for Evolving Cascade Neural Networks

A new learning algorithm for Evolving Cascade Neural Networks (ECNNs) is described. An ECNN starts to learn with one input node and then adding new inputs as well as new hidden neurons evolves it. The trained ECNN has a nearly minimal number of input and hidden neurons as well as connections. The algorithm was successfully applied to classify artifacts and normal segments in clinical electroencephalograms (EEGs). The EEG segments were visually labeled by EEG-viewer. The trained ECNN has correctly classified 96.69% of the testing segments. It is slightly better than a standard fully connected neural network.

Confident Interpretation of Bayesian Decision Tree Ensembles for Clinical Applications

Bayesian averaging (BA) over ensembles of decision models allows evaluation of the uncertainty of decisions that is of crucial importance for safety-critical applications such as medical diagnostics. The interpretability of the ensemble can also give useful information for experts responsible for making reliable decisions. For this reason, decision trees (DTs) are attractive decision models for experts. However, BA over such models makes an ensemble of DTs uninterpretable. In this paper, we present a new approach to probabilistic interpretation of Bayesian DT ensembles. This approach is based on the quantitative evaluation of uncertainty of the DTs, and allows experts to find a DT that provides a high predictive accuracy and confident outcomes. To make the BA over DTs feasible in our experiments, we use a Markov Chain Monte Carlo technique with a reversible jump extension. The results obtained from clinical data show that in terms of predictive accuracy, the proposed method outperforms the maximum a posteriori (MAP) method that has been suggested for interpretation of DT ensembles

Feature Selection for Bayesian Evaluation of Trauma Death Risk

In the last year more than 70,000 people have been brought to the UK hospitals with serious injuries. Each time a clinician has to urgently take a patient through a screening procedure to make a reliable decision on the trauma treatment. Typically, such procedure comprises around 20 tests; however the condition of a trauma patient remains very difficult to be tested properly. What happens if these tests are ambiguously interpreted, and information about the severity of the injury will come misleading? The mistake in a decision can be fatal — using a mild treatment can put a patient at risk of dying from posttraumatic shock, while using an overtreatment can also cause death. How can we reduce the risk of the death caused by unreliable decisions? It has been shown that probabilistic reasoning, based on the Bayesian methodology of averaging over decision models, allows clinicians to evaluate the uncertainty in decision making. Based on this methodology, in this paper we aim at selecting the most important screening tests, keeping a high performance. We assume that the probabilistic reasoning within the Bayesian methodology allows us to discover new relationships between the screening tests and uncertainty in decisions. In practice, selection of the most informative tests can also reduce the cost of a screening procedure in trauma care centers. In our experiments we use the UK Trauma data to compare the efficiency of the proposed technique in terms of the performance. We also compare the uncertainty in decisions in terms of entropy.

The combined technique for detection of artifacts in clinical electroencephalograms of sleeping newborns

In this paper, we describe a new method combining the polynomial neural network and decision tree techniques in order to derive comprehensible classification rules from clinical electroencephalograms (EEGs) recorded from sleeping newborns. These EEGs are heavily corrupted by cardiac, eye movement, muscle, and noise artifacts and, as a consequence, some EEG features are irrelevant to classification problems. Combining the polynomial network and decision tree techniques, we discover comprehensible classification rules while also attempting to keep their classification error down. This technique is shown to outperform a number of commonly used machine learning technique applied to automatically recognize artifacts in the sleep EEGs.

Learning polynomial networks for classification of clinical electroencephalograms

We describe a polynomial network technique developed for learning to classify clinical electroencephalograms (EEGs) presented by noisy features. Using an evolutionary strategy implemented within group method of data handling, we learn classification models which are comprehensively described by sets of short-term polynomials. The polynomial models were learnt to classify the EEGs recorded from Alzheimer and healthy patients and recognize the EEG artifacts. Comparing the performances of our technique and some machine learning methods we conclude that our technique can learn well-suited polynomial models which experts can find easy-to-understand.

A neural-network technique to learn concepts from electroencephalograms.

A new technique is presented developed to learn multi-class concepts from clinical electroencephalograms (EEGs). A desired concept is represented as a neuronal computational model consisting of the input, hidden, and output neurons. In this model the hidden neurons learn independently to classify the EEG segments presented by spectral and statistical features. This technique has been applied to the EEG data recorded from 65 sleeping healthy newborns in order to learn a brain maturation concept of newborns aged between 35 and 51 weeks. The 39,399 and 19,670 segments from these data have been used for learning and testing the concept, respectively. As a result, the concept has correctly classified 80.1% of the testing segments or 87.7% of the 65 records.

Representing classifier confidence in the safety critical domain: an illustration from mortality prediction in trauma cases

This work proposes a novel approach to assessing confidence measures for software classification systems in demanding applications such as those in the safety critical domain. Our focus is the Bayesian framework for developing a model-averaged probabilistic classifier implemented using Markov chain Monte Carlo (MCMC) and where appropriate its reversible jump variant (RJ-MCMC). Within this context we suggest a new technique, building on the reject region idea, to identify areas in feature space that are associated with “unsure” classification predictions. We term such areas “uncertainty envelopes” and they are defined in terms of the full characteristics of the posterior predictive density in different regions of the feature space. We argue this is more informative than use of a traditional reject region which considers only point estimates of predictive probabilities. Results from the method we propose are illustrated on synthetic data and also usefully applied to real life safety critical systems involving medical trauma data.

A Bayesian Model Averaging Methodology for Detecting EEG Artifacts

In this paper we describe a Bayesian Model Averaging (BMA) methodology developed for detecting artifacts in electroencephalograms (EEGs). The EEGs can be heavily corrupted by cardiac, eye movement, muscle and noise artifacts, so that EEG experts need to automatically detect them with a given level of confidence. In theory, the BMA methodology allows experts to evaluate the confidence in decision making most accurately. However, the non- stationary nature of EEGs makes the use of this methodology difficult. In our experiments with the sleep EEGs, the proposed BMA technique is shown to provide a better performance in terms of predictive accuracy.

A Neural-Network Technique for Recognition of Filaments in Solar Images

We describe a new neural-network technique developed for an automated recognition of solar filaments visible in the hydrogen H-alpha line full disk spectroheliograms. This technique deploys the artificial neural network (ANN) with one input and one output neurons and the two hidden neurons associated either with the filament or with background pixels in this fragment. The ANN learns to recognize the filament depicted on a local background from a single image fragment labelled manually. The trained neural network has properly recognized filaments in the testing image fragments depicted on backgrounds with various brightness caused by the atmospherics distortions. Using a parabolic activation function this technique was extended for recognition of multiple solar filaments occasionally appearing in selected fragments.

Classification of newborn EEG maturity with Bayesian averaging over decision trees

Highlights? We analysed the age-related patterns in newborn EEG. ? The use of posterior information about EEG features has improved the assessment. ? The accuracy of the proposed Bayesian assessments slightly outperformed the expert assessments. EEG experts can assess a newborns brain maturity by visual analysis of age-related patterns in sleep EEG. It is highly desirable to make the results of assessment most accurate and reliable. However, the expert analysis is limited in capability to provide the estimate of uncertainty in assessments. Bayesian inference has been shown providing the most accurate estimates of uncertainty by using Markov Chain Monte Carlo (MCMC) integration over the posterior distribution. The use of MCMC enables to approximate the desired distribution by sampling the areas of interests in which the density of distribution is high. In practice, the posterior distribution can be multimodal, and so that the existing MCMC techniques cannot provide the proportional sampling from the areas of interest. The lack of prior information makes MCMC integration more difficult when a model parameter space is large and cannot be explored in detail within a reasonable time. In particular, the lack of information about EEG feature importance can affect the results of Bayesian assessment of EEG maturity. In this paper we explore how the posterior information about EEG feature importance can be used to reduce a negative influence of disproportional sampling on the results of Bayesian assessment. We found that the MCMC integration tends to oversample the areas in which a model parameter space includes one or more features, the importance of which counted in terms of their posterior use is low. Using this finding, we proposed to cure the results of MCMC integration and then described the results of testing the proposed method on a set of sleep EEG recordings.