Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2022 Jul 23:2022:7833516.
doi: 10.1155/2022/7833516. eCollection 2022.

Deep Learning-Based Networks for Detecting Anomalies in Chest X-Rays

Malek Badr  1   2   3 , Shaha Al-Otaibi  4 , Nazik Alturki  4 , Tanvir Abir  5
Affiliations

Deep Learning-Based Networks for Detecting Anomalies in Chest X-Rays

Malek Badr et al. Biomed Res Int. .

Retraction in

Abstract

X-ray images aid medical professionals in the diagnosis and detection of pathologies. They are critical, for example, in the diagnosis of pneumonia, the detection of masses, and, more recently, the detection of COVID-19-related conditions. The chest X-ray is one of the first imaging tests performed when pathology is suspected because it is one of the most accessible radiological examinations. Deep learning-based neural networks, particularly convolutional neural networks, have exploded in popularity in recent years and have become indispensable tools for image classification. Transfer learning approaches, in particular, have enabled the use of previously trained networks' knowledge, eliminating the need for large data sets and lowering the high computational costs associated with this type of network. This research focuses on using deep learning-based neural networks to detect anomalies in chest X-rays. Different convolutional network-based approaches are investigated using the ChestX-ray14 database, which contains over 100,000 X-ray images with labels relating to 14 different pathologies, and different classification objectives are evaluated. Starting with the pretrained networks VGG19, ResNet50, and Inceptionv3, networks based on transfer learning are implemented, with different schemes for the classification stage and data augmentation. Similarly, an ad hoc architecture is proposed and evaluated without transfer learning for the classification objective with more examples. The results show that transfer learning produces acceptable results in most of the tested cases, indicating that it is a viable first step for using deep networks when there are not enough labeled images, which is a common problem when working with medical images. The ad hoc network, on the other hand, demonstrated good generalization with data augmentation and an acceptable accuracy value. The findings suggest that using convolutional neural networks with and without transfer learning to design classifiers for detecting pathologies in chest X-rays is a good idea.

PubMed Disclaimer

Conflict of interest statement

There is no potential conflict of interest in our paper, and all authors have seen the manuscript and approved to submit to your journal.

Figures

Figure 1
Figure 1
Chest X-ray images, with their respective labels, belonging to the ChestX-ray14 data set.
Figure 2
Figure 2
General architecture of the model based on transfer learning used.
Figure 3
Figure 3
Comparison between a typical CNN block and a ResNet residual block.
Figure 4
Figure 4
Illustrative figure of the dropout technique.
Figure 5
Figure 5
Dynamic data augmentation.
Figure 6
Figure 6
Simple hold-out partition of the data set.
Figure 7
Figure 7
Subsets made up of pure pathology and “without finding.”
Figure 8
Figure 8
Average accuracy results for architecture #1.
Figure 9
Figure 9
Comparison of results obtained by filtering by orientation and without filtering.
Figure 10
Figure 10
Average accuracy without and with data augmentation for the data sets made up of one pathology and “no finding.”
Figure 11
Figure 11
Accuracy vs. epoch for the CNN without transfer learning, trained with the data set composed of the labels “no finding” and “finding.”
See this image and copyright information in PMC

References

    1. Stead M., Bower M., Brinkmann B., Warren C., Worrell G. A. In Epilepsy: The Intersection of Neurosciences, Biology, Mathematics, Engineering, and Physics . CRC Press; 2016. Large-Scale Electrophysiology: Acquisition, Storage and Analysis. - DOI
    1. Wang X., Peng Y., Lu L., Lu Z., Bagheri M., Summers R. M. ChestX-Ray8: hospital-scale chest X-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases. In Proceedings of the IEEE conference on computer vision and pattern recognition; 2017; Honolulu. pp. 2097–2106. - DOI
    1. Chauhan R., Ghanshala K. K., Joshi R. C. Convolutional neural network (CNN) for image detection and recognition. In 2018 First International Conference on Secure Cyber Computing and Communication (ICSCCC); 2018; Jalandhar, India. pp. 278–282. - DOI
    1. Madiajagan M., Raj S. In Deep learning and parallel computing environment for bioengineering systems . Academic Press; 2019. Parallel computing, graphics processing unit (GPU) and new hardware for deep learning in computational intelligence research; pp. 1–15. - DOI
    1. Alazzam M. B., Tayyib N., Alshawwa S. Z., Ahmed M. Nursing care systematization with case-based reasoning and artificial intelligence. Journal of Healthcare Engineering . 2022;2022 doi: 10.1155/2022/1959371. - DOI - PMC - PubMed

Publication types