Piotr Wais

Artificial immune system for medical data classification

The article presents application of artificial immune algorithms in classification of vectorized medical data sets. Artificial immune network was created and trained for the purpose of arterial blood gasometry parameters (pH, pCO2, pO2, HCO3) classification. Training data originates from the Infant Intensive Care Unit of the Polish – American Institute of Pediatry, Collegium Medicum, Jagiellonian University in Cracow.

An Artificial Immune Algorithms Apply to Pre-processing Signals

Over the past few decades there has been a growing interest in the use of biology as a source of inspiration for solving computational problems. The motivation of this field is primarily to extract useful metaphors from natural biological systems, in order to create effective computational solutions to complex problems in a wide range of domain areas. The more notable developments have been the neural networks inspired by the working of the brain, and the evolutionary algorithms inspired by neo-Darwinian theory of evolution. This paper presents the theory of an immune network model, and it tries to apply to solve signal classification problems.

Arterial Flows in Bronchopulmonary Dysplasia Prediction

The paper presents Bronchopulmonary Dysplasia, BPD, prediction for extremely premature infants after their first week of life using LR (Logit Regression). Presented models give accuracy up to 84.6% using only three independent variables and 81.7% with two of them. That novelty was possible to achieve thanks to unique use of arterial flows measurements, which are not a common clinical practice. That original data was collected thanks to the Neonatal Intensive Care Unit of The Department of Pediatrics at Jagiellonian University Medical College. The main pulmonary artery (MPA) and patent ductus arteriosus (PDA) flows were considered as predictors. Beyond classic statistic significance analysis and LR forecast paper presents some other results and discussion, which give an outlook on possible repeatability of results and its quality on some other’s patients data set.

Prediction of Arterial Blood Gases Values in Premature Infants with Respiratory Disorders

Arterial blood gases sampling (ABG) is a method for acquiring neonatal patients’ acid-base status. Variations of blood gasometry parameters values over time can be modelled using multi-layer artificial neural networks (ANNs). Accurate predictions of future levels of blood gases can be useful in supporting therapeutic decision making. In the paper several models of ANN are trained using growing numbers of feature vectors and assessment is made about the influence of input matrix size on the accuracy of ANNs’ prediction capabilities.

Bronchopulmonary Dysplasia Prediction Using Naive Bayes Classifier

The paper presents BPD (Bronchopulmonary Dysplasia) prediction for extremely premature infants after their first week of life. In contrast to the most works where LR (Logit Regression) is used, the naive Bayes classifier was proposed. Data was collected thanks to the Neonatal Intensive Care Unit of The Department of Pediatrics at Jagiellonian University Medical College and includes 109 patients with birth weight less than or equal to 1500 g. Fourteen different features were considered and all \(2^{14}\) of theirs combinations were analyzed. This paper also includes an accuracy and its deviation comparison with other prediction methods. It was possible because the calculations were performed on the very same data, which was used in previous works presenting LR and SVM forecasts.

Statistical Method for the Problem of Bronchopulmonary Dysplasia Classification in Pre-mature Infants

The problem of data classification with a statistical method is presented in the paper. Described classification method enables calculation of probability of disease incidence. A case of disease incidence is described with two parameters expressed in real numbers. The case can belong to a known set of cases where the disease occurred or to the set where the disease did not occur. A method for calculating probability with which a given case belongs to the set labeled as “1” or “0” is proposed. Source data used in the paper come from medical databases and are original. The algorithm of the method was checked on clinical cases. Correlation method was used for generating respective statistics. The calculated correlation at a level of 0.8 is indicative of disease occurrence, whereas the correlation coefficient at a level of 0.0 is indicative of the lack of disease. This property is used in the classification algorithm. It is frequent in the clinical practice that we have one test case and we try to determine whether or not that case describes symptoms of liability to the disease. Classification is related with the occurrence of Bronchopulmonary dysplasia, which is analyzed in a 3 to 4 week period preceding the disease incidence.

Arterial Blood Gases Forecast Optimization by Artificial Neural Network Method

Arterial blood gas sampling represents the gold standard method for acquiring patients’ acid-base status. It is proposed that blood gas values could be measured using arterialized earlobe blood samples. Pulse oximetry plus transcutaneous carbon dioxide measurement is an alternative method of obtaining similar information as well. Since dynamics of biochemical changes occurring in the blood is an individual feature which changes during the healing process authors proposed forecast models developed using artificial neural networks. The networks are trained with data vectors containing short term (72 h) history windows of four blood gasometry parameters. Several different optimization algorithms are used in the training phase to create a set of models from which the best prediction model is then selected.

Predictive analysis of blood gasometry parameters related to the infants respiration insufficiency

The paper presents application of artificial immune system in time series prediction of the medical data. Prediction mechanism used in the work is basing on the paradigm stating that in the immune system during the response there exist not only antigene – antibody connections but also antigene – antigene connections, which role is control of antibodies activity. Moreover in the work learning mechanism of the artificial immune network, and results of carried out tests are presented.

D09: Classification of vectorized medical data sets using artificial immune algorithms

Abstract The article presents application of artificial immune algorithms in classification of vectorized medical data sets. Artificial immune network was created and trained for the purpose of arterial blood gasometry parameters (pH, pCO2, pO2, HCO3) classification. Training data originate from the Infant Intensive Care Unit of the Polish – American Institute of Pediatry, Collegium Medicum, Jagiellonian University in Cracow.