2016 年 24 巻 2 号 p. 26-32
Improving the reliability of a physician's diagnosis also has a major impact on the patient. Studies related to liver diseases have been carried out since the early days of the Mahalanobis-Taguchi (MT) system, and in the MT methods that use inverse matrices, even when the true value is unknown, diagnostic decisions have been made with a standardized Mahalanobis distance scale by use of explanatory variables. When the true value is unknown, however, it is not easy to determine and study the discriminatory capability of the obtained distances or the suitability of their sensitivity. In the present study the correlation coefficients of the unit space were corrected by use of a discount coefficient method. The correction is shown to increase the distance values of the data and improve the S/N ratio of the discriminations. In the cases studied here, as compared with the conventional method, the detection sensitivity of the Mahalanobis distance for liver patients was improved by approximately a factor of two.