Histopathological diagnosis of brain tumours with the help of a computer: mathematical fundaments and practical...

With the help of a personal computer we have subjected 1684 tumours of the central and peripheral nervous systems to a discriminant analysis to test whether the different brain tumours can be classified. By ascribing a coefficient value which belongs to a linear function to the objective criteria for each tumour these can be ordered in groups, and a classification is then possible. This has been termed discriminant analysis. Fifty histological characteristics were the criteria taken into consideration for each tumour. By using Bayes' formula with a correction for binomial distribution we obtained a correct diagnosis in 98% of the cases studied. The differential diagnosis with its percentage of probabilities is given by the computer. The mathematical fundaments are given. The valuability and precision of this method are discussed.