Retrieving physiological features in decision support systems

The Artificial Intelligence in Medical Applications (AIM) project at the Department of Computer Science and Electronics (IDE) officially started in january 2002 and is mainly externally funded. A three year grant from Kunskap och Kompetens Stiftelsen (KKS) 1 was granted in mid 2001 for the funding of two doctoral students in Artificial Intelligence (AI). Additional funding was matched by PBM StressMedicine 2 as they would benefit from the research. The department provides additional funding by providing workplaces, supervisors etc. The supervisors are Peter Funk and Björn lisper. The area of interest for this research is to capture knowledge with AI based methods for the use in decision support systems. The knowledge in this case pertain to a few experts in a special branch of psychophysiological medicine[13] where practitioners and clinicians use physiological measurements to diagnose patients with stress related dysfunctions[11, 17, 18]. The treatment is often based on some sort of biofeedback training[7, 9]. This is our definition of psychophysiology in the remainder of the document. The project is divided into two parts, one per doctoral student. The first part, which is the part this doctoral thesis proposal is based upon, handles physiological parameters , i.e., measurements from patients[4, 7]. They have to be examined for important information, such features clinicians find useful in their diagnosis. The approach for finding these features is to analyse the raw signals, thus signal theory is a essential part of this part of the project. The first part can be used as a stand-alone decision support system or be used as a pre-processor to the second part of the AIM project. Additional help for clinicians is provided by the second part of the project. This part identifies important sequences of pre-classified signals by clustering sequences to known patterns. This gives a more complete analysis of the measurements as it explains the measurements in a more familiar enviroment. The research has lead to the development of a new classification method for the identification of important features in physiological parameters [16] in medical support systems. The classification method is implemented in a decision support system, HR3Modul[15], for diagnosing respiratory sinus arrhythmia (RSA) [18]. It is in general difficult to find patterns within biomedical signals, i.e. physiological time series. This also applies to our domain of interest. The RSA domain has also the non-favourable attribute of beeing unkown in some parts. It is not widely known how …