Machine Learning in Stepwise Diagnostic Process

The diagnoses of many diseases have become increasingly complex. Many results, obtained from tests with substantial imperfections , must be integrated into a diagnostic conclusion about the probability of disease in a given patient. A practical approach to this problem is to estimate the pretest probability of disease, and the sensitivity and speciicity of diierent diagnostic tests. With this information, test results can be analyzed by sequential use of Bayes' theorem of conditional probability. The calculated post-test probability is then used as a pre-test probability for the next test. This results in a series of test where each test is performed independently and its results may be interpreted with or without any knowledge of the other test results. By using Machine Learning techniques, this process can be almost completely automated. The computer can learn from previously diagnosed patients and apply the learned knowledge in new patients. Diierent Machine Learning methods can be used for evaluation of each test result. They may assist the physician as a powerful tool for assistance in pre-test probability estimation , interpretation of the individual test results and in the nal decision making. The presented approach has been successfully evaluated in practice in the problem of clinical diagnosis of ischaemic heart disease.