Dynamic Selection Classification for Chest X-Ray Images

In this article, we investigate how well combinations of classifiers work on chest X-ray images when using classical extracted features and deep learning generated features. We employed four methods for classical feature extraction and seven types of features generated with some well-known pre-trained neural networks. A dynamic selection of classifiers method based on accuracy is used for mixing these eleven classification results. The classification results are computed with a linear SVM (Support Vector Machine) using these eleven sets of features. For experiments, we employed the Kermany dataset that contains chest X-ray images of the lungs. In all situations, the combination of classifiers improved the individual results (up to 7% in the best case).