Pattern Classification

Abstract The first computerized medical decision support systems were based on pattern classification techniques. Early applications were focused on the diagnosis of disease. The process involved determination of parameters that were potentially useful in classification of cases into two or more categories. The parameters, denoted features, were organized as vector components. Numerous learning algorithms were developed that were trained on available data both for supervised learning in which correct classification of cases was known and unsupervised learning in which the classification was unknown and occasionally even the number of classes was unknown. The models were very successful and multiplied rapidly. Applications of pattern recognition in decision making came to be known as pattern classification, whereas pattern recognition is more often associated with recognition of patterns in images. The basic techniques are the same, with the differences originating in the feature identification phase. Neural networks are special types of pattern classification algorithms that use architecture based on biological nervous systems.