Classifying High-Dimensional Lung Cancer Data Using an Improved Deep Learning Model

Abstract

     Anomalous cell growth, known as cancer, can cause tumors, weaken the immune system, and cause fatal complications. Early detection of cancer makes treatment easier and contributes to lower mortality rates. Gene expression data contributes crucially to the classification of cancer in its early stages. Accurately classifying cancer is a complex problem because gene expression data is high-dimensional compared to the limited number of samples. This research proposes an improved Deep Learning model that integrates both Supervised Autoencoder (SAE) and Convolutional Neural Network (CNN), named as (SAE-CNN), to address this problem. Specifically, the SAE model was first developed to reduce the dimensionality of the dataset by identifying the most influential genes that will be considered as the main input data for the proposed effective CNN cancer classification model. The proposed cancer classification model is applied to several lung cancer gene expression microarray datasets that include distinct classes of cancer. The experimental results demonstrate that the proposed model outperforms existing state-of-the-art methods and proves its effectiveness in cancer classification, although the size of high-dimensional gene expression data is limited. Where Lung_Harvard2, Lung_Michigan, and Lung_Ontario datasets achieved a maximum accuracy of 100%, while Lung_Adenocarcinoma dataset achieved 90.48%. Moreover, the results demonstrate that the proposed model has the potential to help treat cancer and increase survival rates in the future.