Implementation and validation of a computerized expert system sleep scoring is proposed. Such a system would increase productivity of sleep laboratories, and increase the quality while reducing the cost of sleep diagnostic procedures. Phase I research will focus on the basic algorithms for automated sleep stage scoring. The algorithms are based on Bayesian decision theory and follow the steps of the human expert in the decision making process when scoring sleep by the Rechtschaffen-Kales standard. Likelihood estimates of the suggested and alternative decisions will be used to quantify confidence in suggested scores. The system will flag low confidence scores to be revised by human interaction. Initial tests show that the level of agreement between machine and human is similar to that between two human scorers. Software tools will be developed to test the effect of the parameters of the algorithms on stage score decisions. An initial data base of training and test sets of normal and apneic sleep recordings will be established and the algorithms will be optimized and tested on this data. Phase II will optimize parameters and validate the system on larger data sets for other diagnostic and age groups, develop on-line, real-time capabilities, and extend the decision model to facilitate further research in sleep medicine.
| Norman, R G; Pal, I; Stewart, C et al. (2000) Interobserver agreement among sleep scorers from different centers in a large dataset. Sleep 23:901-8 |
