Deep learning-based analysis of COVID-19 x-ray images: a comparative study of colourmap
Background: With the emergence of the SARS-CoV-2 virus late in 2019, the world’s healthcare system has been severely affected by the COVID-19 pandemic, necessitating the need for quick and effective actions to reduce its extensive effects. Chest X-ray (CXR) imaging is critical for accurate assessmen...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Institute for Health Management, Ministry of Health Malaysia
2023
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/111539/1/111539_Deep%20learning-based%20analysis.pdf http://irep.iium.edu.my/111539/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Background: With the emergence of the SARS-CoV-2 virus late in 2019, the world’s healthcare system has been severely affected by the COVID-19 pandemic, necessitating the need for quick and effective actions to reduce its extensive effects. Chest X-ray (CXR) imaging is critical for accurate assessment, displaying intricate lung structural abnormalities, including ground-glass opacities, consolidation, and bilateral infiltrates in COVID-19 patients. The objective of this study was to examine the comparison between grayscale and 16 colourmap images in terms of their efficacy in COVID-19 detection when used with the DarkNet-53 deep learning architecture. Methodology: We conducted an experiment with a dataset of 9,665 CXRs, consisting of 7,134 normal images and 2,531 COVID-19 images, in order to train deep learning architectures. An additional dataset of 4,143 CXRs, with 3,058 normal and 1,085 COVID-19 images, was used for independent testing. The images underwent pre-processing and were split into grayscale and 16 colourmap images for individual examination. The COVID-19 detection task was fine-tuned on DarkNet-53, a deep learning architecture, with standard data augmentation techniques applied to grayscale and 16 colourmap images. Results: The DarkNet-53 deep learning architecture demonstrated verifying results based on the X-ray image utilised. The bone colourmap achieved the highest accuracy (0.985) and sensitivity (0.952) scores, while the grayscale, pink, and summer colourmaps demonstrated the greatest specificity (0.998). Conclusion: Our study highlights the importance of choosing the right type of X-ray image in association with deep learning architecture for CXR COVID-19 detection. These outcomes have important consequences for automating and upgrading CXR analysis, aiding in the exact detection of COVID-19 and respiratory health issues, and eventually benefiting patient care and outcomes. |
|---|