The Difficulties of Decision Trees in the Diagnostic of Acute Abdominal Pain

摘要

Decision support systems that help physicians are becoming a very important part of medical decision making. One of the most viable among them are decision trees, already successfully used for many medical decision making purposes. Transparent and straightforward representation of accumulated knowledge and fast algorithms made decision trees what they are today: one of the most often used symbolic machine learning approaches. This paper describes a new decision tree approach MtDeciT to the clinical field of acute abdominal pain. Acute abdominal pain manifests itself in considerable diagnostic error rates and negative outcomes, despite improvements by using imaging technology (e.g. ultrasound) and special laboratory investigations. It is a frequent problem with the necessity for urgent management decisions. Studies in this area have reported diagnostic accuracy of 60% by the initial examiner and of 80% by the final examiner. Different standard induction techniques (e.g. ID3, NewID, C4.5) were tested on the same prospective database with an overall accuracy in a range between 40% and 48% on the test set.