IBM Type 704 Medical Diagnosis Program

The basic input to the type 704 computer is simply the presence or absence of those characteristics (symptoms or results of a laboratory test) of a set of medical cases which are considered to be pertinent by an expert in that particular field. Pairwise and over-all similarity of both the cases and the symptoms are defined so that geometrically the cases, considered as points in a suitable space, form several clumps, thus establishing a natural classification. The mathematical model is based on the assumption of equal-likeliness of the characteristics or symptoms so that in order that the final results be significant, a sufficiently large number of characteristics are initially assumed and the system is reduced by repetitively deleting the least significant characteristic, thus inducing weighting of the characteristics, since more emphasis is placed on those characteristics which comprehend one another. When the set of characteristics is fixed, the computer is able to determine the case or cases which in an over-all sense best represent the characteristics. By interpreting pairwise similarity as probabilities, it is possible to predict how likely it is for a case with given symptoms (or tests) to exist on the basis of the classification the computer produces.