Clinical data analysis with the use of artificial neural networks (ANN) and principal component analysis (PCA) of patients with endometrial carcinoma

Abstract

Aim

Satisfactory performance of modern data processing methods, namely principal component analysis (PCA) and artificial neural network (ANN) analysis, has been demonstrated in the prediction of the results of surgical treatment for endometrial carcinoma.

Materials/Methods

The data from 121 patients treated and observed in one oncology unit was retrospectively evaluated. 26 subject and treatment variables were determined for each patient. A matrix of 121×26 data points was subjected to PCA and ANN processing.

Results

The properly trained ANN was used to predict whether patients belonged to the group of those who survived, or to the group of those who did not survive, a 5-year period. It was found that the prognostic capability of the ANN, regarding the tested set of patients, was very high. Additionally, using the principal component analysis (PCA) method, two principal components, PC1 and PC2 were extracted and accounted, cumulatively, for 23% of the variance in the data analyzed. An apparent clustering of the variables and a clear cut clustering of the patients was observed, which has been interpreted in terms of similarity, or dissimilarity, of the variables and of the patients.

Conclusions

It has been concluded that ANN analysis offers a promising alternative to established methods for the statistical analysis of multivariate data in cancer patients. Also, PCA has been recommended as a new and promising alternative to classical regression analysis of multivariate clinical data. By means of PCA, practically useful systematic information may be extracted from large sets of data, which is otherwise hardly interpretable in comprehensive physical terms. Such information can be of value for general prognosis and for making appropriate adjustments in treatment.