Kernel-based Support Vector Machine classifiers for early detection of myocardial infarction

In this paper, we describe the development of kernel-based Support Vector Machine (SVM) classifiers to aid the early diagnosis of acute myocardial infarction (AMI). In particular, we have to recognize if a chest pain, complained by the patient, may be considered the sign of a myocardial infarction or it is the evidence of some other causes. This is a quite difficult medical decision problem, since chest pain is characterized by low specificity (typical values between 30% and 40%) as a symptom associated with myocardial infarction. Moreover, in order to make an objective and accurate diagnosis, the physician has to evaluate a large set of data coming from the patient. These aspects motivated the use of machine learning methodologies, with the aim to support the physician and increase the quality of the diagnostic decision. To this end, we formulated the medical decision problem as a supervised binary classification problem (AMI class and not AMI class), by developing a training set with 242 cases (130 in the AMI class and 112 in the not AMI class), each case characterized by a set of 105 features. We also considered a feature selection procedure, by selecting 25 of the 105 features. By the framework of generalized SVM model, we tested and validated the behavior of three kernel functions: Polynomial, Gaussian and Laplacian. By running a 10-fold cross validation procedure, the performance of the best tested classifier was 97.5%. By the same 10-fold cross validation procedure, we tested linear and quadratic discriminant analysis classifiers, with testing correctness of 86.8% and 94%, respectively. The numerical results demonstrate the effectiveness and robustness of the proposed approaches for solving the relevant medical decision making problem.