A Novel Spectral Feature Selection Method Based on Binary Genetic Algorithm for Efficient Detection of Endometrial and Ovarian Cancers: Preliminary Results

Endometrial and ovarian cancers are the most common gynecological cancers. Early and accurate diagnosis of both diseases is essential for minimizing mortality rate as well as treatment costs. In this study, we propose a novel methodology for efficient detection of endometrial and ovarian cancers. The proposed approach is based on processing urine-based Fourier Transform infrared spectroscopy data by using machine learning powered by evolutionary feature selection. We propose a novel spectral feature selection method based on a binary genetic algorithm, which is combined with a number of well-known machine learning methods and applied on spectroscopy data obtained from urine samples. Our results show that the presented approach offers superior performance over other methods in the literature for the same purpose. It can therefore efficiently be used for accurate and automated detection of endometrial and ovarian cancers by utilizing a very small set of spectral features, which provides a significant benefit regarding physical implementation of the proposed pipeline as a biosensor kit.