Artificial Neural Networks and Medical Decision Making Renato M.E. Sabbatini Center for Biomedical Informatics, State University of Campinas, Campinas SP 13083-970, Brazil. Artificial neural networks, or connectionist systems, are being increasingly used to represent and to processs information by means of networks of interconnected processing elements, similar to neurones. Several emerging global properties of connectionist systems, such as associative memory, distributed parallel processing, learning, etc.; have favored its applications in a large variety of tasks involving pattern classification and recognition. Connectionist systems constitute a new and interesting paradigm for the area of Artificial Intelligence in Medicine, and have found applications in processing and interpretation of biological signals and images, decision support systems, etc., substituting advantageously several conventional techniques based on statistical classification and logical systems. We review here the uses of ANNs in the building of medical decision making systems. As a new Artificial Intelligence approach, the connectionist paradigm is able to accept and to process several kinds of formal representation of knowledge, such as feature vectors, probabilities, hierarchical structures, semantic and belief networks, and even production rules. As general pattern classifiers, ANNs can be used as efficient tools in areas where traditional, heuristics-oriented, logical-based expert systems have dominated, such as medical diagnosis and therapy. Thus, there has been an increase in the number of applications devoted to clinical decision-support using ANNs. They have already found way in many medical decision support areas, as diverse as neurology, psychiatry, ophtalmology, endocrinoloy, cardiology, gastroenterology, dermatology, neonatology, surgery, orthopedics, anesthesiology, oncology, radiology, radiation therapy, clinical pathology, nursing and alternative medicine. These systems generally work in the following way: a fixed set of indicators, such as symptoms, signs, results of laboratory tests, etc., are presented to the input layer of a network. The vector portraying the corresponding diagnosis, prognosis or therapeutic recommendation (which may be multiple) is applied to the output layer. ANNs are especially useful, in contrast to conventional AI approaches, in representing and treating uncertain, probabilistic, approximate, fuzzy or noisy concepts and data, either at the input or the output level. The network is then trained to associate input and output patterns, based on a number of training (usually real) clinical cases derived from a database. After achieving convergence and stability, the remaining (test) clinical cases in the database is used to verify the performance of the network in classifying correctly known patterns. After this, it is considered able (or not) to diagnose unknown cases. Most of the research papers usually compare the network's performance to other computer-based decision-support systems, or to human experts' evaluations. We will exemplify practical uses of ANNs in medical prognosis, successfuly developed in our laboratory, showing applications to outcome prediction for critical patients under intensive care, and for patients with dilative myocardial disease, using supervised learning procedures with multilayer perceptrons