Medicine: Applications in Machine Learning
share
Health care has been an important issue in computer science since the 1960s. In addition to databases storing patient records, library resources (e.g., PubMed, a service of the U.S. National Library of Medicine that includes over 16 million citations from journals for biomedical articles back to the 1950s), administrative and financial systems, more sophisticated support of health care has been the aim of artificial intelligence (AI) from the very beginning on. Starting with expert systems which abstract laboratory findings and other vital parameters of a patient before they heuristically classify the patient into one of the modeled diagnoses (Shortliffe, 1976), knowledge acquisition was discovered to be the bottleneck of systems for the automatic medical diagnosis. Machine learning came into play as a means of knowledge acquisition. Learning rules for (medical) expert systems focused on the heuristic classification step within expert systems. Given conveniently abstracted measurements of the patient’s state, the classification was learned in terms of rules or decision trees. Since the early days, the use of machine learning for health care progressed in two ways: The abstraction of measurements of a patient’s vital parameters is a learning task in its own right. Diverse kinds of data are to be handled: laboratory data, online measurements at the bedside, x-rays or other imaging data, genetic data,... Machine learning is confronted with a diversity of representations for the examples. Diagnosis is just one task in which physicians are to be supported. There are many more tasks which machine learning can ease. In intensive care, the addressee of the learning results can be a machine, e.g., the respirator. Financing health care and planning the medical resources (e.g., for a predicted epidemia) are yet another important issue. Machine learning is placed in a diversity of medical tasks. The urgent need for sophisticated support of health care follows from reports which estimate up to 100,000 deaths in the USA each year due to medical error (Kohn, Corrigan, & Donaldson, 2000).
READ FULL TEXT
