Author/Editor		Grabec, Igor; Ferkolj, Ivan; Grabec, Daša; Grošelj, Dušan
Title		Prediction of clinical response to treatment of Chron's disease by using RBFN
Type		članek
Source		Dyn Contin Discrete Impuls Syst
Vol. and No.		Letnik 14, št. Suppl 1
Publication year		2007
Volume		str. 602-7
Language		eng
Abstract		This paper concerns the prediction of patient response to treatment of Crohn's disease with the drug infliximab. As an optimal predictor, a normalized radial basis function neural network is utilized. Information used in the prediction is based on joint data from clinical parameters and response indicators observed in a test group of patients. In the presented example, the network utilizes given oral parameters from selected patients to predict the response to drug administration. In the prediction algorithm, the similarity between the given parameters and those in the database is calculated. If is then used as a weight by which the response of a patient from the test group is predicted. The method thus resembles the prediction performed by a physician based upon comparison of a treated patient with previously tested ones. Prediction quality is estimated from the discrepancy between predicted and observed response data. Prediction quality corresponding to particular clinical parameters provides for their ordering and selection of an optimal set of parameters that together yield the maximal quality O.63 and - 80% coincidence between predicted and observed response categories.
Descriptors		CROHN DISEASE TUMOR NECROSIS FACTOR ANTIBODIES, MONOCLONAL NEURAL NETWORKS (COMPUTER)