Early Breast Cancer Detection from Blood Plasma Using Hubness-Aware Adaptive Neural Network with Hybrid Feature Selection

Abstract

Breast cancer is a complex and heterogeneous disease prevalent amongst women leading to mortality. Breast cancer diagnosis at the early stages of tumour development can increase the survival rate of women suffering from breast cancer. The most prevalent mammogram method for early screening of breast cancer results in a greater number of false positives for breast cancer detection in young women. Latest research in the field of genomics suggests that microRNAs (miRNAs) of circulating tumour cells, and gene expressions of circulating tumour cells in blood plasma are good candidate for the detection of breast cancer during early stages of cancer. Hence, it is proposed to develop a diagnostic machine learning model for early detection and effective treatment from the gene expression values of miRNA. Since gene expression data are complex and high dimensional, a hybrid feature selection is performed using molecular pathway information, and gene ontology information related to breast cancer and also most informative genes are selected using Laplacian score feature ranking. For the classification of samples into tumour and normal samples, a novel improved artificial neural network (ANN), namely hubness-aware adaptive neural network (HAANN) with adaptive learning rate is implemented. A novel strategy based on hubness score is adapted to adjust the learning rate of the neural network dynamically. Performance analysis of the proposed machine learning shows an improved performance compared to the traditional classification algorithms for intrinsic high-dimensional data.