KARDIO: A Study in Deep and Qualitative Knowledge for Expert SystemsThis book is the first detailed account of the development of a complex and successful expert system based on deep and qualitative knowledge. It shows how the qualitative modeling approach, using logic based representations and machine learning techniques, can be used to construct knowledge bases whose complexity is far beyond the capability of traditional, dialogue based techniques of knowledge acquisition. The relevant techniques are demonstrated in full detail in the building of Kardio, a medical expert system model of the human heart designed for the diagnosis of cardiac arrhythmias. Kardio's performance is estimated by cardiologists to be equivalent to that of a specialist of internal medicine (not a cardiologist) who is highly skilled in the reading of ECG recordings, and it can be used as a diagnostic tool in ECG interpretation. It may also be used for instruction in electrocardiography. The authors show how the model was compiled, by means of qualitative simulation and machine learning tools, into various representations that are suited for particular expert tasks. They investigate a hierarchical organization of a qualitative model and outline an experiment whereby the construction of a deep model is automated by means of machine learning techniques. The book contains a complete model of the electrical system of the heart that can be used to further development in this area of applications. Ivan Bratko, author of Prolog Programming for Artificial Intelligence, is a professor of computer science at E. Kardelj University and leads the AI laboratory at the Jozef Stefan Institute in Ljubljana, Yugoslavia. Igor Mozetic and Nada Lavrac are researchers at the institute. |
Contents
Introduction | 1 |
Qualitative Model of the Heart | 53 |
Model Interpretation and Derivation | 95 |
Copyright | |
5 other sections not shown
Other editions - View all
Common terms and phrases
abnormal absent aeb V quiet after_P_some_QRS_miss algorithm Arrs atr_focus atrial attributes AV node av_block AV_cond av_conduct av_junction avb1 avb3 blown Bratko bulb bundle branches cardiac arrhythmias circuit clauses combined arrhythmias compressed rules conduction corresponding ECG debugger deep model defined derived Disj Disj0 disjunctive dominant QRS dominant_PR ECG descriptions ECG features ect_vent_focus ectopic beats ectopic_beats ectopic_QRS(1 expert system Figure function heart components heart disorders his_bundle hypothesis independent_P_QRS inductive learning irregular KARDIO knowledge base Lavrač learning program logic machine learning meaningless V after_P_always_QRS meaningless V after_QRS_is_P Michalski mob2 Mozetič NEWGEM normal V wide_LBBB occasional impulses P_wave permanent impulses positive examples possible prediction rules Prolog QRS complex qualitative model qualitative simulation Rate rate_of_P rate_of_QRS Rate0 Ratel reduced arrhythmia-ECG base reg_vent_focus regular relation_P_QRS after_P_always_QRS representation rhythm_QRS sa_node shortened V after_QRS_is_P simple arrhythmias sinus rhythm surface knowledge tachycardia under_60 values ventricles ventricular wide_LBBB V wide_non_specific wide_RBBB V wide_LBBB zero zero_60