Selecting the next neuroradiological investigation with the help of a computer
G. H. du Boulay, Vivien Price · British Journal of Radiology · 1971
The use of a computer to assist the diagnosis of intracranial tumours was described by the present authors in 1968. Subsequent work has been concerned with improving the accuracy of the diagnostic program by experimental weighting of more important radiological signs (du Boulay, Popple, and Price 1969) and with the addition of proven cases to the computer's file of data. From the beginning of these studies a second program has been in operation. Its purpose has been to choose, by comparison with past proven cases, what should be the “next test” (in previous publications spoken of as the “next best test”). This next test, carotid angiography for instance, will be seen to be “best” if it can be shown statistically by comparison with previous patients that it is the most likely test to discriminate between the three most probable diagnoses. Theory of the method incorporated in the second program at present is based on the three diagnoses (denoted by i1, i2 and i3) which have been given the highest probability by the first program. For each of these values of i and for each investigation, J, not yet made on the new patient the computer calculates: where j = a basic test, i.e. an individual sign such as pineal displacement, obtained by doing a comprehensive test, e.g. plain film. Nij = total number of patients with the ith type of tumour on whom the jth test was made.