抄録
Computer diagnosis or numerical diagnosis begins from the criticism on the feature of the medical system at present. Medical information at present is given as a system of deductive description and mostly qualitative ones. As a fact, however, disease profiles in one disease have numerous faces and one disease profile corresponds to many diseases. Therefore the process of diagnostic thinking can not be deterministic, but probabilistic. In numerical diagnosis a disease profile is dealt as a multi-variate and evaluated by the statistical measure called likelihood. In the field of the estimation of the name of the diseases the following will be not worthy : researches on cerebellar tumors by Takahashi et al. (Univ. of Tokyo), brain tumors in cortex by Domae et al. (ibid), apoplexia by Fukui et al. (ibid), traumatic injury of the head by Sano et al. (ibid), heart diseases by Dohi et al. (ibid), acute abdomen dy Shikata (Sumida Hospital), liver diseases by Kosaka et al. (Okayama Univ.) and so on. In the field of pattern recognition of laboratory data electrocardiogram by Okajima et al. (Nagoya Univ.), by Kimura (Nippon Med. College), by Matsuo (Tohoku Univ.) and by others, phonocardiogram by Yoshimura (Jikei Med. College) and by Machii (Univ. of Tokyo), findings of gastrocamera by Hirokado et al. (Kyushu Univ.) and so on. In most cases numerical diagnosis gives more excellent result than doctors'.
In most of these researches multivariate statistics are applied. Boolean algebra and Bayesian probability, introduced by American researchers Ledley and Lusted, revealed that these are not adequate as a measure of diagnostic inference, either theoretically or experimentally.
Though there is no doubt on the effectiveness of numerical diagnosis, probability itself has no final meaning. It is accepted as one of the elements of medical decision theory. Another field to be developed in future is the estimation of the prognosis.
