Determination of risk factors using Nonlinear Principal Component Analysis in patients with breast tumour

Abstract

Objective: Breast cancer, which is the most common among women in the world and constitutes approximately 30% of all cancers, takes places near the top among the diseases that threaten women's health. The purpose of this study is to determine the risk factors in patients with breast tumours using nonlinear principal component analysis.
 Materials and Methods: During the application process, a data set of 569 (357 benign, 212 malign) patients with breast tumours was used. To find independent features, the data set was reduced to two dimensions via nonlinear principal component analysis. The results were evaluated by comparing the success of the method with the ROC curve.
 Results: The cut-off values for the radius, perimeter, area, smoothness and texture of the tumour were 14.19, 656.10, 0.09, 2.87 and 0.11, respectively. The sensitivity of the current values according to the results of ROC analysis was determined as 84% for radius, 80% for perimeter, 86% for the area and 94% for texture. It is seen that the method has an overall success of over 80% in detecting malignant tumours.
 Conclusions: It is hoped that this method, which is used to reveal risk factors and identify distinctive features in breast tumours, will reduce medical costs and provide a second opinion to physicians. In terms of decision making, it is predicted that the method can recognize malignant tumours and reduce the need for unnecessary biopsy for benign tumours.