lntomouonal roumal of Medical lnfofmatics ELSEVIER International Journal of Medical Informatics 45 (1997) 129-132 Abstracts The utilization of neural networks in medical informatics Sebesta V Institute of Computer Science, Acudemy of Sciences of‘ the Czech Republic, Pod vod&enskou veli 2, I8207 Prague, Czech Republic Keywords: Neural network; Diagnostics; Prediction The development of a multilayer neural network may be divided into several steps: (1) problem formulation-determi- nation of the task, which has to be solved (classification-pre- diction-data compression-diagnostics); (2) determination of the network topology (a larger than needed network may be designed and trained); (3) specification of learning patterns and relevant markers (sufficient number of patterns with known results and acceptable distribution among all classes, in which we want to classify); (4) determination of synaptic weights and other parameters of neurons (the problem of global optimization has to be solved, although no conditions for the existence of a solution are usually known); (5) evalua- tion of the generalization ability and (6) modifications by pruning the previously designed neural network. The utilization of neural networks has the following advan- tages: 0 no knowledge of the exact algorithms for solution is neces- sary; l very fast operation in the phase of utilization; 0 possibility of future adaptation, when it is necessary: it also has the following disadvantages; 0 huge time consumption during the phase of learning; 0 difficulties in the determination of topology; 0 difficulties in the creation of the training set, which must represent all possible classes; 0 possibility of false results. with only limited tools for the recognition of this fact. Some examples of Neural Networks Utilization in medicine: 0 General or specialised diagnostics (classification) myocar- dial infarction, coronary artery disease, PET scans of Alzheimer’s disease. 0 Evalution of time-series (Signal Processing) evaluation of ST-T segments in ECG, evaluation of alpha and beta activity in EEG. 0 Prediction of time-series, prediction of survival time, opti- mization of operation time, optimization of drug consump- tion. 0 Decision support system for the decision if a drug may be used, for optimal determination of the time and type of drugs, for the optimal therapy choice. l Picture processing, discovery of abnormal cells according to the recognized attributes, e.g. geometrical measures, discov- ery of tumours on the basis of different graphical transfor- mations of X-ray pictures, etc. Computer simulation in clinical practice Svacina S, Haas T, Hovorka, R 3rd Medical Clinic, 1st Medical Faculty Charles University, 128 21 Prague, Katerinska 32, Czech Republic Keywords: Computer simulation; Clinical medicine; Psychol- ogy Computer simulation is a cognitive method. Computer models are mostly used to analyze dynamics of physiological systems. Let us pose the question: what is the role of computer simula- tion in clinical practice? We sum up this role according to our experience with several original and adapted models. Com- puter simulation in clinical practice can be used for these reason (followed by examples): 0 Quantifying of dynamic clinical findings (minimal model of glycoregulation in intravenous glucose tolerance tests [I], insulin receptor binding calculation using binding model [2], individual glycation parameters calculating in our model of hemoglobin and protein glycation [3]). Some simple vari- able is used instead of curves to describe dynamics in a concrete patient. l Investigation in clinical physiology (e.g. hemoglobin glyca- tion [3] or insulin absorption research [4]). 0 Model based device parameter setting (we have experience with artificial pancreas-patient interaction modelling and model based parameter setting [4]). This approach can be also used in dialysis or cardiac pacemaker parameters. 0 Presentation of knowledge in consultation systems (see our example [5] of a model based consultation system for insulin therapy). 0 Prediction and prognosis of diseases (risk calculation of type 2 diabetes uses predicting model [6]). We are using also another model based calculation for malnutrition classifica- tion and risk calculations for surgery. 0 Teaching of clinical decision making (insulin consultation system can be used for teaching [5]). 1386-5056/97/$17.00 0 1997 Elsevier Science Ireland Ltd. All rights reserved. P11SOO2O-7101(97)00041-X