Expert knowledge for computerized ECG interpretation

textabstractIn this study, two main questions are addressed: (1) Can the time consuming and cumbersome development and refinement of (heuristic) ECG classifiers be alleviated, and (2) Is it possible to increase diagnostic performance of ECG computer programs by combining knowledge from multiple sources? Chapters 2 and 3 are of an introductory character. In Chapter 2, the measurement part of MEANS is described and evaluated. This research largely depends on the earlier work of Talman [11]. In Chapter 3, different methods of diagnostic ECG classification are described and their pros and cons discussed. The issue is raised whether or not the ECG should be classified using as much prior information as possible, and our position is made clear. The first question~ how to ease the transfer of cardiological knowledge into computer algorithms, is addressed in Chapters 4 and 5. The development and refinement of heuristic ECG classifiers is impeded by two problems: (1) It generally requires a computer expert to translate the cardiologist's reasoning into computer language without the average cardiologist being able to verify whether his diagnostic intentions were properly realized, and (2) The classifiers are often so complex as to obscure insight into their doings when a particular case is processed by the classification program. To circumvent these problems. we developed a dedicated language. DTL (Decision Tree Language), and an interpreter and compiler of that language. In Chapter 4, a comprehensive description of the DTL environment is given. In Chapter 5, the use of the environment to optimize MEANS, following a procedure of stepwise refmement, is described The second question, whether it is feasible to combine knowledge from multiple sources in order to increase diagnostic performance of an ECG computer program, is explored from several perspectives in Chapters 6 tlrrough