Donna Hudson

Research Interests

Computer-assisted medical decision making, neural network modeling, medical time series using chaos theory.

Research Summary

Our group's research involves development of techniques for computer-assisted biomedical decision-making, including knowledge-based systems, neural network models, and hybrid systems. These approaches encompass methods for analysis of non-textual data such as images and signals. Our group has developed a hybrid system that includes EMERGE, a knowledge-based system, HYPERNET, a non-statistical neural network model that can accept a variety of variable types, and CATS, a chaotic-based model that is used to summarize variability in biomedical times series. The knowledge-based system inference engine relies on techniques for approximate reasoning and was initially implemented for analysis of chest pain in the emergency room. Current theoretical work focuses on the addition of more sophisticated reasoning techniques, including consequential reasoning in which potential consequences of an action are included in the decision process. The neural network model uses a potential function approach based on the class of Cohen orthogonal functions. Resulting nonlinear decision surfaces have been shown to be effective in classification of data sets from numerous medical applications over the last 15 years. New theoretical methods of chaos theory have been used to develop summary measures of biomedical time series that produce both graphical and numerical indicators for the degree of chaos in the series. Extensive work has been done in the application of this method to the analysis of R-R intervals obtained from 24-hr Holter tapes in patients with congestive heart failure. These measures have shown promise in identification of heart failure and also in the differentiation among different categories of heart disease. Current theoretical work is focused on preprocessing techniques such as wavelet analysis that can be used to detect patterns at differing amplitudes, a particularly useful technique in the analysis of electroencephalograms. Current theoretical work on the combined hybrid system is directed toward the inclusion of intelligent agents that allow seamless transitions among the different components of the system. Currently, a large project is underway to develop a comprehensive decision model for diagnosis and staging of Alzheimer's Disease that will include expert-supplied rules, neural network models based on patient cases, and analysis of EEG data.

Selected Publications

Hudson, D.L., Cohen, M.E., Kramer, M., Meechem, M., Inclusion of Signal Analysis in a Hybrid Medical Decision Support System, Methods of Information in Medicine, 43:79-82, 2004. 

Hudson, D.L., Cohen, M.E., The Role of Networks and Artificial Intelligence in Nanotechnology Design and Analysis, Cell and Molecular Biology, 50(3):297-300, 2004.

Hudson, D.L., Cohen, M.E., Kramer, M., Szeri, A., Chang, F.L.Diagnostic Implications of EEG Analysis in Patients with Dementia, IEEE EMBS Proceedings on Neural Engineering, 2:629-632, 2005.

Hudson, D.L., Medical Expert Systems, Encyclopedia of Biomedical Engineering, John Wiley and Sons, 2006.

Hudson, D.L., Pattern Recognition, Encyclopedia of Biomedical Engineering, John Wiley and Sons, 2006.