Avtor/Urednik		Kukar, Matjaž; Grošelj, Ciril
Naslov		Uporaba cenovno občutljivega učenja za ocenjevanje predtestnih in izboljševanje potestnih verjetnosti v diagnostiki ishemične bolezni srca
Prevedeni naslov		Cost sensitive learning for estimation of pretest probalities and improvement of posttest probalities in the diagnosis of ischaemic heart disease
Tip		članek
Vir		Inform Med Slov Print Ed
Vol. in št.		Letnik 5, št. 1-2
Leto izdaje		1998
Obseg		str. 77-82
Jezik		slo
Abstrakt		The main goal of the physician in daily practice is to make the right diagnosis for deciding the best therapy. However, on many occasions the physician must make decisions with some degree of uncertaity. During the process of clinical diagnosis, the maximal certainty about the patient's disease is attempted using tools such as the history, physical examination and laboratory studies. One way by which the degree of uncertainty of the diagnosis can be quantified is by expressing it in terms of probabilities. Important parameters of this approach are pretest probability of the disease, the choice of diagnostic test and posttest probability of th disease. In the usual diagnostic process lots of potentially useful medical data (results of previous tests) remains unused. This data can be used for the calculation of pretest probabilities from personal experience or bibliography. In our work we applied Machine Learning algorithms for learning pretest probability estimations from the training set of pre-diagnosed patients. Our experiments show that this approach gives better results than the results obtained by physicians. By using the cost-sensitive Machine Learning methods for classification we can also adjust the diagnostic process to suit the momentary needs (i.e., higher sensitivity or specificity of the diagnostic test.).
Deskriptorji		CORONARY DISEASE DIAGNOSIS, COMPUTER-ASSISTED DECISION MAKING, COMPUTER-ASSISTED SENSITIVITY AND SPECIFICITY