Classification of chest CT scans using convolutional autoencoders and local binary patterns

Abstract

Caused by an uncontrolled growth of cells in the lung, lung cancer is one of the leading causes of death around the world. This study introduces a novel computer-aided diagnosis system to classify chest CT-scans into two classes i.e. malignant and non-malignant. We train a convolutional autoencoder to extract deep features from input images and we combine the result with descriptors generated using local binary patterns. The concatenated vector is fed into a one-dimensional convolutional neural network to perform the classification task. The proposed model outperformed all other models in the literature while being tested on a subset of the IQ-OTHNCCD lung cancer dataset.