This paper describes some results of statistical analysis and automatic diagnosis of respiratory diseases, particularly lung cancer, on a digital computer.
We consider a
n-dimensional vector, called a symptom vector, that fixes one patient at time T
o :
X= (
X1,
X2………
Xn) where, if the patient shows the
ith symptom,
Xi=1, if so not,
Xi=0. These
nelements are regarded as coordinates in a
n-dimensional vector space. Statistical analysis was aimed at relating certain subspaces of the vector space with the pathological categories.
The hyperplane by which the vector space was separated into two subspaces, each subspace corresponding to pathological category, was chosen for the statistical analysis mainly because of its conceptual and mathematic simplicity. The basic conception of the optimum separation is based on the minimax principle.
The above mentioned method seems to be superior to a usual linear discriminant function, on two following points :
1) It is possible to use this method in the case that the covariance matrix in each. population is different.
2) This method may be used, not only in the case that each variable is continuous, but in the case that each variable is discrete.
Consequently, this paper shows that two pathological groups can be classified by means of the expression of linear combination of each variable in the case that medical information, is discrete.
As for medical information, the authors used 18 items of patient's history and 24 findings of broncho graphy respectively.
With each sample, the method has been satisfactory. However, demonstration of clinical evaluation will require additional investigation.
View full abstract