抄録
The studies on the automatic diagnosis of respiratory diseases, particularly lung cancer, were attempted by means of a digital electronic computer. The authors devised the coding of posteroanterior chest roentgen findings, becaus these findings were considered to be an essential key to respiratory disease diagnosis. The roentgenograms were divided into minimal shadow units and the coding of such properties of each unit as localization, shape, size, density, margin homogeneity, number, distribution and concentricity was made. This coding seems to be superior to the routine expression of roentgen findings, because the possibility of misreading of the shadows due to examiners' inexperience or misunderstanding due to different interpretation between individual examiners is much reduced and, therefore, the reproducibility of this coded findings is also improved. Five standard types of reontgen findings of lung cancer and pulmonary tuberculosis were coded by our method and memorized by the digital computer as binary numbers. The roentgenograms of 66 cases of lung cancer and 184 cases of pulmonary tuberculosis were also coded by our method and expressed by binary numbers, then typed into the computer, b&ng compared with the stored standard types of the diseases to determine whether these cases were idential with any stored types of diseases or not. 53 per cent of the primary lung cancer were typed out as identical with more than one standard types of both diseases and 37 per cent as identical with more than one standard types of only the lung cancer, thus 90 per cent of the cases of the primary lung cancer were picked up as possible lung cancer. In the cases of the metastatic lung cancer the percentage was not so high. 88.6 per cent of the cases of the pulmonary tuberculosis were also picked up as possible pulmonary tuberculosis. From these results, the proposed method of the coding can be considered adequate to the automatic diagnosis system of the respiratory diseases by means of digital electronic computers.
