Erstellung und Evaluierung von Entscheidungsbäumen

The focus of this work was the investigation of the following
questions in current research: How can one define an objectively
verifiable procedure for the veterinarian during the diagnostic
process and how can this procedure be represented at best? Is it
possible to validate this procedure in a general way? Are the
answers to these previous questions appropriate to lead to more
effective and efficient diagnoses in veterinary medicine? Symptoms
which are relevant for the practical work in the small animal
medicine branches dermatology, cardiology and neurology, were
divided into sub-groups(rule outs) following a fixed concept which
is based on clearly defined criteria. By utilizing this approach, a
procedure for the veterinarian is proposed. This systematic
framework of the necessary knowledge for finding a diagnosis was
visualized by the author in the form of decision trees which were
saved on power point slides. Subsequently, these decision trees
were validated qualitatively by interviews and discussions with
professors and senior lecturers of the Clinic for Small Animal
Medicine of the LMU Munich. In a next step, this diagnostic support
tool was evaluated by the author with students utilizing four tests
and a questionnaire. The evaluation of these tests showed that the
students who were allowed to use the decision trees, reached higher
scores and needed less time to solve the questions. These results
support the thesis that the decision trees are appropriate to
increase the efficiency and effectiveness and, in consequence, are
a positive contribution to solving clinical cases. Compared to
other diagnostic support tools (some of these have only a limited
practicability as support tools), the proposed decision trees can
convey detailed knowledge in categorized, structured and visualized
form to veterinarians and students of veterinary medicine, so that
this tool represents a contribution to finding a diagnosis in a
more structured and objective way. In consequence, diagnostic
decisions will be safer, which is especially important for
inexperienced veterinarians and veterinary students. The generated
decision trees can be utilized as a basis for further
investigations of symptoms and should stimulate scientific
discussions in the research field "diagnosis". In this context, the
creation of a database is conceivable, which could present the
contents and more detailed information in a multimedia-based way.