Home > Published Issues > 2022 > Volume 10, No. 2, June 2022 >
JOIG 2022 Vol.10(2): 56-63
doi: 10.18178/joig.10.2.56-63

Automatic Out-of-Distribution Detection Methods for Improving the Deep Learning Classification of Pulmonary X-ray Images

Andrey A. Dovganich 1, Alexander V. Khvostikov 1, Yakov A. Pchelintsev 1, Andrey A. Krylov 1, Yong Ding 2, and Mylene C. Q. Farias 3
1. Department of Computational Mathematics and Cybernetics, Lomonosov Moscow State University, Moscow, Russia
2. Zhejiang University, Hangzhou, China
3. University of Brasilia, Brasilia, Brazil

Abstract—In this paper, we investigated the possibility of using medical differential criteria to determine the level of radiation in X-ray images of the lungs. We developed a new method for automatic determination and calculation the number of visible vertebrae in the pulmonary X-ray images and proposed a system of automatic out-of-distribution detection that can be used together with deep learning-based systems of pulmonary X-ray image analysis, in particular with the task of tuberculosis detection. The proposed method and system were evaluated using three X-ray lung datasets (Montgomery County chest X-ray dataset, Shenzhen chest X-ray dataset and Tuberculosis X-ray TBX11K dataset). We demonstrated that using the proposed system of out-of-distribution detection allows to enhance the tuberculosis classification results up to 1.3% using the same classification model. We also showed that the proposed system allows to automatically train a composite model which considers X-ray radiation level of the image, which is more effective compared to the traditional one-part model.

Index Terms—X-ray images, deep learning, quality assessment, out-of-distribution detection, detection of vertebrae

Cite: Andrey A. Dovganich, Alexander V. Khvostikov, Yakov A. Pchelintsev, Andrey A. Krylov, Yong Ding, and Mylene C. Q. Farias, "Automatic Out-of-Distribution Detection Methods for Improving the Deep Learning Classification of Pulmonary X-ray Images," Journal of Image and Graphics, Vol. 10, No. 2, pp. 56-63, June 2022.

Copyright © 2022 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.