Vaishnavi Kardale, Bioinformatics Centre, Savitribai Phule Pune University
The COVID-19 pandemic has been around for a year and a half, causing the death of thousands and infecting millions worldwide. Due to its highly contagious nature, the most effective way to keep the spread under control is to keep social distance and contact tracing. Hence, early diagnosis has become crucial to counter the spread.
RT-PCR (reverse transcription-polymerase chain reaction) is the standard procedure that is popularly used for COVID-19 testing. However, due to the low accuracy of the RT-PCR and the limited availability of test kits, it is challenging to detect every individual that has been detected with COVID-19. This has necessitated the development of an alternate testing method for faster and more reliable COVID diagnosis.
Alternate COVID testing using X-rays
As most COVID-19 positive patients also suffer from pneumonia, radiological examinations could help detect the disease. Computed tomography (CT) scans and X-rays can alternatively be used for real-time COVID detection. Previously deep learning-based methods have been used for COVID-19 detection using CT scan images. However, the machinery to conduct CT scans is more expensive and takes considerably more time than X-ray imaging. Compared to CT, X-rays can speed up the screening and have become a preferred method for disease diagnosis.
Oh et al. have proposed a convolutional neural network trained on an open-source dataset COVID-net for the screening of COVID-19 using chest X-ray (CXR). The CXR shows similar pathological information between COVID-19 and pneumonia. However, these latent features can be misclassified by the hyperplane learned from limited training data. The uncertainty in COVID-19 detection is still a major challenge for existing deep networks and this problem is further elevated due to the presence of noise in the training dataset.
To address the above concerns, researchers from Zhejiang University (Hangzhou, China), proposed a novel deep network architecture called RCoNet for robust COVID detection.
What is RCoNet?
The RCoNet contains three modules:
- Deformable mutual Information Maximization (DeIM)
The DeIM allows the model to learn the discriminative and compact features. CNN employs predefined grids that fail to detect and classify occluded and deformed images. Deformable Convolution Network (DCN) is more advanced in the sense that the grid points can be moved. These new points are augmented by a learnable offset (offset is a bias value that is used while training the model). RCoNet employs DCN so that disentangled spatial features can be recognized.
- Mixed High-order Moment Feature (MHMF)
The MHMF module helps to better understand and characterize the feature distribution in medical imaging. The MHMF is benefitted by the use of a mix of high-order moment statistics. This also helps in reducing the negative effect of noise.
- Multiexpert Uncertainty-aware Learning (MUL)
The MUL improves the prediction accuracy by creating multiple parallel dropout networks. Each network can be treated as an expert. This gives us multiple expert-based diagnoses just like clinical practice. The prediction accuracy is quantified by obtaining the variance in prediction across different experts.
Conclusion
Dong et al. numerically validated that RCoNet, trained on publicly available COVIDx and CXR images of noisy settings, outperformed existing methods. The researchers suggest the use of the above three modules in other frameworks for different tasks as well.
Also read: Trabecular bone texture analysis in assessing osteoarthritis
Reference:
Dong, S., Yang, Q., Fu, Y., Tian, M., & Zhuo, C. (2021). RCoNet: Deformable Mutual Information Maximization and High-Order Uncertainty-Aware Learning for Robust COVID-19 Detection. IEEE Transactions on Neural Networks and Learning Systems, 32(8), 3401–3411. https://doi.org/10.1109/TNNLS.2021.3086570
- The Corrosion Prediction from the Corrosion Product Performance
- Nitrogen Resilience in Waterlogged Soybean plants
- Cell Senescence in Type II Diabetes: Therapeutic Potential
- Transgene-Free Canker-Resistant Citrus sinensis with Cas12/RNP
- AI Literacy in Early Childhood Education: Challenges and Opportunities
Author info:
Vaishnavi Kardale is a master’s student at the Bioinformatics Centre, Savitribai Phule University. She is interested in protein folding mechanisms and wants to study them further.
Publications:
https://bioxone.in/news/worldnews/global-warming-may-reduce-the-spread-of-dengue/
https://bioxone.in/news/worldnews/comeback-of-tuberculosis-but-its-drug-resistant-now/
https://bioxone.in/news/worldnews/a-drug-to-reduce-covid-infection-by-99/
https://bioxone.in/news/worldnews/artificial-intelligence-ai-for-efficient-covid-testing/
https://bioxone.in/news/worldnews/deephbv-a-machine-learning-tool-to-aid-in-hepatitis-b-integration-site-detection/
One thought on “RCoNet: Diagnosing COVID-19 using chest X-rays”