Lung Disease Classification in Chest X-Rays with Feature Fusion via Stacked Convolutional Denoising Autoencoders

Deep Learning recently has made a breakthrough and tremendous advancement in medical imaging domain. While training supervised deep learning models require large number of annotated data. This is costlier, challenging and time-consuming and requires domain expertise. Nevertheless, these challenges can be tackled with models trained in unsupervised manner., that requires no annotations. Stacked convolutional denoising autoencoders are unsupervised models with convolutional layers that reconstruct the original input data that requires no annotated data with random perturbations added to the original input data. The ability to combine information from various sources in enhancing classification performance is showing promising results. In this paper, we show unsupervised models with stacked convolutional denoising autoencoders for extracting salient features from input un annotated data and fine tuning it for supervised classification of lung diseases on chest X-ray images with feature fusion.