Ovarian Cancer Detection based on Dimensionality Reduction Techniques and Genetic Algorithm

TL;DR

This study compares feature selection and classification methods for ovarian cancer detection using serum mass spectra, finding genetic algorithms combined with neural networks achieve perfect accuracy, outperforming PCA with LDA.

Contribution

The paper demonstrates that genetic algorithms combined with neural networks outperform PCA and LDA in ovarian cancer detection from serum proteomic data.

Findings

GA with neural networks achieved 100% accuracy

PCA with LDA achieved 93.02% accuracy

GA is more efficient for feature selection in this context

Abstract

In this research, we have two serum SELDI (surface-enhanced laser desorption and ionization) mass spectra (MS) datasets to be used to select features amongst them to identify proteomic cancerous serums from normal serums. Features selection techniques have been applied and classification techniques have been applied as well. Amongst the features selection techniques we have chosen to evaluate the performance of PCA (Principal Component Analysis ) and GA (Genetic algorithm), and amongst the classification techniques we have chosen the LDA (Linear Discriminant Analysis) and Neural networks so as to evaluate the ability of the selected features in identifying the cancerous patterns. Results were obtained for two combinations of features selection techniques and classification techniques, the first one was PCA+(t-test) technique for features selection and LDA for accuracy tracking yielded an accuracy of 93.0233 % , the other one was genetic algorithm and neural network yielded an accuracy of 100%. So, we conclude that GA is more efficient for features selection and hence for cancerous patterns detection than PCA technique.