Pretrained Convolutional Neural Networks for Cancer Genome Classification

Abstract

Deep learning techniques, particularly convolutional neural networks (CNNs), have proved their success and popularity recently in many fields, especially distinguishing and analyzing medical diseases. Motivated by this direction, our work attempts for the first time to investigate the application of a state-of-the-art deep learning technique on genomic sequences to classify tumours of different classes. The novelty of our approach lies in the application of the popular pre-trained AlexNet on an image version of the RNA-Sequence data. Our methodology demonstrated an outstanding performance with good sensitivity results of 98.3%, 94.1%, 96.6%, 100%, and 100% for selected types of breast, colon, kidney, lung and prostate cancers respectively. The outcome of this work is expected to provide a new direction for genomics data classification and designing accurate automated diagnosis tools.