Using decision tree algorithms as a basis for a heart sound diagnosis decision support system

Decision trees algorithms were used with promising results in various critical problems, concerning heart sound diagnosis. In general this diagnostic problem can be divided in many sub problems, each one dealing either with one morphological characteristic of the heart sound or with difficult to distinguish heart diseases. The sub problems of the discrimination of aortic stenosis from mitral regurgitation and the discrimination between the second heart sound split, opening snap and third heart sound, are used as case studies. Using signal-processing methods, we extracted the heart sound feature vector. Relevance analysis was performed using the uncertainty coefficient. Then for each heart sound diagnosis sub problem, a specific decision tree (DT) was constructed. decision tree pruning was also investigated. Finally, a general decision support system architecture for the heart sound diagnosis problem, is proposed. The partial diagnosis, given by these DT, can be combined using arbitration rules to give the final diagnosis. These rules can be implemented by another DT, or can be based on different methods, algorithms, or even on expert knowledge. All these can lead to an integrated decision support system architecture for heart sound diagnosis.