Abstract—32 Hemodynamic parameters (HDPs) derived from sphygmogram (SPG) and 5 physiological parameters (PPs) are widely used for cardiovascular diseases (CVDs) detection. All these parameters are divided to groups naturally for diagnostic usage, which conforms to doctors’ clinical diagnosis procedure. The number and type of HDP&PP in groups are varied according to detecting different CVDs, in another word the grouping is disease-oriented, which leads to a bottleneck problem: how to construct a hierarchical high-efficient classifier to diagnose CVDs based on grouped HDPs & PPs? To tackle such a formidable problem, a hierarchical classifier HPSVM based on support vector machine with probabilistic outputs (PSVM) is proposed. Such formed classifier has good generalization ability with unique global solution even with small training dataset. It also conforms to the doctors’ hierarchical diagnosis procedure and reduces the deduction complexity with high diagnostic accuracy. Site-measured datasets obtained from Beijing Changping Chinese Medicine Hospital are used for testing and the results verify the prospect of this technology with higher than 90% accuracy in detecting three typical and frequently encountered CVDs.

Index Terms—Classification and regression tree, disease-oriented grouping, hierarchical classifier, quadratic programming, support vector machine with probabilistic outputs.

Mubo Chen, Binbin Fu, Jiali ma, and Mingchui Dong are with the Department of Electrical and Computer Engineering, faculty of Science and Technology, University of Macau, Macau S.A.R., China (e-mail: chenmubocs@gmail.com, ariespleo51@gmail.com, jimohanqiu@hotmail.com, mcdong@umac.mo).
Taichun Tang is with the Department of Software Engineering, Faculty of Science and Technology, University of Macau, Macau S.A.R., China (e-mail: tangtai009@163.com).

Cite: Mubo Chen, Binbin Fu, Taichun Tang, Jiali Ma, and Mingchui Dong, "Hierarchical Probabilistic Support Vector Machine for Detecting Cardiovascular Diseases," International Journal of Bioscience, Biochemistry and Bioinformatics vol. 4, no. 5, pp. 340-344, 2014.