Improving Uncertainty Estimation With Semi-Supervised Deep Learning for COVID-19 Detection Using Chest X-Ray Images

IEEE Access

Luis Oala¹  Ezequiel Lopez-Rubio²  Jorge Rodriguez-Capitan²  Saul Calderon-Ramirez³  Shengxiang Yang³  Simon Colreavy-Donnelly³  David A. Elizondo³  Armaghan Moemeni⁴  Miguel A. Molina-Cabello⁵  Manuel Jimenez-Navarro⁵ 

[1] Artificial Intelligence Department, XAI Group, Fraunhofer Heinrich Hertz Institute, Berlin, Germany;CIBERCV, Hospital Universitario Virgen de la Victoria, M&x00E1;School of Computer Science and Informatics, De Montfort University, Leicester, U.K.;School of Computer Science, University of Nottingham, Nottingham, U.K.;laga, Spain;

关键词: Uncertainty estimation;    Coronavirus;    Covid-19;    chest x-ray;    computer aided diagnosis;    semi-supervised deep learning;   

DOI  :  10.1109/ACCESS.2021.3085418

来源: DOAJ

【 摘 要 】

In this work we implement a COVID-19 infection detection system based on chest X-ray images with uncertainty estimation. Uncertainty estimation is vital for safe usage of computer aided diagnosis tools in medical applications. Model estimations with high uncertainty should be carefully analyzed by a trained radiologist. We aim to improve uncertainty estimations using unlabelled data through the MixMatch semi-supervised framework. We test popular uncertainty estimation approaches, comprising Softmax scores, Monte-Carlo dropout and deterministic uncertainty quantification. To compare the reliability of the uncertainty estimates, we propose the usage of the Jensen-Shannon distance between the uncertainty distributions of correct and incorrect estimations. This metric is statistically relevant, unlike most previously used metrics, which often ignore the distribution of the uncertainty estimations. Our test results show a significant improvement in uncertainty estimates when using unlabelled data. The best results are obtained with the use of the Monte Carlo dropout method.

【 授权许可】

Unknown