ABSTRACT
In the previous report, the authors described a method obtaining numerical values of 'standard deviation (SD)' and deviation coefficient (DC) from printed out data of a brain CT slice picking up the basal ganglia. In the report, using these values, a possibility of numerical discrimination between normal and abnormal was also discussed. In the present investigation, it was attempted to apply admissible linear discriminant analysis to numerical diagnosis using SD value and DC value. From fifty normal scans and fifty abnormal ones of proved pathology, the admissible linear discriminant function, as below, was extracted, its misdiscriminant rate being 7.6% theroretically. S=-0.4248(SD)-6.4709 (DC) S larger than or equal to -11.447; normal, S < -11.447; abnormal. By this analysis, 98% of abnormal scans and 92% of normal scans were correctly judged without referring to analogue display and actual misdiscriminant rate was 5.0%. Furtermore, for the purpose of confirming reappearance of this result, another twenty normal scans and twenty abnormal scans were investigated by this method. Only one normal scan was judged uncorrectly and all abnormal scans were judged correctly. Accordingly misdiscriminant rate of this sample was 2.5%. It would be emphasized that this method was useful for screeing examination by CT, not only because the total discriminant rate was high, but also false negative rate was relatively low.
