Source-linked AI summary
Fusion of convolution neural network, support vector machine and Sobel filter for accurate detection of COVID-19 patients using X-ray images
Danial Sharifrazi, Roohallah Alizadehsani, Mohamad Roshanzamir, Javad Hassannataj Joloudari, Afshin Shoeibi, Mahboobeh Jafari, Sadiq Hussain, Zahra Alizadeh Sani, Fereshteh Hasanzadeh, Fahime Khozeimeh, Abbas Khosravi, Saeid Nahavandi, Maryam Panahiazar, Assef Zare, Sheikh Mohammed Shariful Islam, U Rajendra Acharya
TL;DR
Early COVID-19 recognition is important because the disease is highly contagious and RT-PCR has limited sensitivity for early suspected cases. This paper proposes a Sobel-filtered CNN with an SVM classifier for X-ray detection, reporting strong performance across private and public databases. The authors also identify high computational cost and limited input data as limitations.
Problem
COVID-19 requires rapid diagnosis because it is highly contagious, while RT-PCR sensitivity is not high enough for early recognition of suspected patients.
Method
The method applies Sobel filtering to X-ray images, extracts features with a CNN, and performs classification with an SVM.
Results
The CNN-SVM+Sobel method performed better than other tested combinations across the six public databases and the study databases.
Takeaways & Limitations
Sobel filtering improved the performance of the CNN-based approaches, and the method was tested on both a newly collected database and six public databases.
Takeaways & Limitations
The authors identify high computational cost among deep-learning algorithms and limited input data as weaknesses of the proposed algorithm.
Abstract
from arXiv · showhide
The coronavirus (COVID-19) is currently the most common contagious disease which is prevalent all over the world. The main challenge of this disease is the primary diagnosis to prevent secondary infections and its spread from one person to another. Therefore, it is essential to use an automatic diagnosis system along with clinical procedures for the rapid diagnosis of COVID-19 to prevent its spread. Artificial intelligence techniques using computed tomography (CT) images of the lungs and chest radiography have the potential to obtain high diagnostic performance for Covid-19 diagnosis. In this study, a fusion of convolutional neural network (CNN), support vector machine (SVM), and Sobel filter is proposed to detect COVID-19 using X-ray images. A new X-ray image dataset was collected and subjected to high pass filter using a Sobel filter to obtain the edges of the images. Then these images are fed to CNN deep learning model followed by SVM classifier with ten-fold cross validation strategy. This method is designed so that it can learn with not many data. Our results show that the proposed CNN-SVM with Sobel filtering (CNN-SVM+Sobel) achieved the highest classification accuracy of 99.02% in accurate detection of COVID-19. It showed that using Sobel filter can improve the performance of CNN. Unlike most of the other researches, this method does not use a pre-trained network. We have also validated our developed model using six public databases and obtained the highest performance. Hence, our developed model is ready for clinical application
1. Introduction
COVID-19 presents a major global healthcare challenge, while RT-PCR has limited sensitivity for early recognition. The paper motivates deep-learning-based X-ray analysis and highlights its dataset and validation contributions.
- COVID-19 is highly contagious and can cause respiratory, cardiovascular, and multiorgan complications, making it a global healthcare challenge.
- RT-PCR is generally used for confirmation, but its sensitivity is insufficient for early recognition of suspected patients.
- The study uses a new private database, data augmentation, six public databases, and Sobel filtering to develop and evaluate its detection approach.
- Deep learning can automatically extract features for medical-image analysis, addressing a key challenge in conventional machine-learning methods.
- The paper reviews deep-learning studies using X-ray and CT images, noting that almost all used pre-trained networks.
2. CADS Based COVID-19 Diagnosis Using 2D-CNN
The proposed CAD system combines a 2D-CNN with either sigmoid or SVM classification for COVID-19 detection from X-ray images. It uses augmentation, preprocessing, convolutional feature extraction, and binary evaluation metrics.
- The proposed 2D-CNN uses convolutional, max-pooling, and fully connected layers, with SVM replacing sigmoid as an alternative final classifier.
- The pipeline augments X-ray data, resizes and normalizes images, extracts features through convolutional layers, and classifies them using sigmoid or SVM.
- The dataset contains 333 chest X-ray images: 77 from COVID-19 patients and 256 from normal subjects, assessed by three radiologists.
- Data augmentation increases the dataset from 333 to 1332 images using width shifts, height shifts, and rotations.
- Performance measures: Performance is evaluated with accuracy and related classification measures based on true-positive, false-negative, true-negative, and false-positive outcomes.
3. Results and Discussion
Across private, augmented, and six public databases, the proposed CNN-SVM and Sobel-filter combinations were evaluated with ten-fold cross-validation. The authors report that Sobel filtering improved CNN performance and that CNN-SVM+Sobel performed best across the tested databases.
- Evaluation setup: The study evaluated CNN-Sigmoid, CNN-SVM, and their Sobel-filter combinations on private and augmented COVID-19 X-ray databases using ten-fold cross-validation.The private database contained 1332 images after augmentation from 333 original images.
- Private and augmented databases: Sobel filtering significantly improved both CNN-Sigmoid and CNN-SVM approaches for COVID-19 detection.The authors attribute additional improvement to using SVM as a more robust classifier.
- Public databases: Across six public databases, Sobel filtering had a positive effect in all tests, while CNN-SVM+Sobel performed better than the other combinations.The authors also report that CNN-Sigmoid+Sobel outperformed the remaining combinations in the public-database comparisons.
- Comparison with prior work: The proposed method was compared with other research methods and was reported to perform better on the authors’ database.The comparison is reported in Table 6 for COVID-19 detection using X-ray images from different private databases.
- Private and augmented databases: CNN-SVM+Sobel performed better than the other method combinations on the private and augmented databases.Figures 16 and 17 compare the proposed methods on the original private database and its augmented version.
- Limitations: The authors identify high computational cost and limited input data as limitations of the proposed method.These limitations are stated in the paper’s list of disadvantages.
4. Conclusion
The study develops a COVID-19 detection system using Sobel-filtered X-ray images, CNN feature extraction, and SVM classification. It achieves 99.02% accuracy on the authors’ dataset and highest detection accuracy across six public databases.
- Conclusion: The system applies a Sobel filter to X-ray images before CNN processing and SVM classification.The authors used 333 X-ray images from Omid Hospital, Tehran, comprising 77 COVID-19 and 256 normal cases.
- Conclusion: 99.02% accuracy was achieved for COVID-19 detection on the authors’ dataset.
- Conclusion: The model yielded the highest detection accuracy using six public databases, supporting its reported robustness.
- Conclusion: The authors propose extending the model to other chest diseases, including cancer, pneumonia, cystic fibrosis, infection, and COPD.