Covid-19: Automatic detection from X-Ray images utilizing Transfer Learning with Convolutional Neural Networks

Abstract: In this study, a dataset of X-Ray images from patients with common pneumonia, Covid-19, and normal incidents was utilized for the automatic detection of the Coronavirus. The aim of the study is to evaluate the performance of state-of-the-art Convolutional Neural Network architectures proposed over recent years for medical image classification. Specifically, the procedure called transfer learning was adopted. With transfer learning, the detection of various abnormalities in small medical image datasets is an achievable target, often yielding remarkable results. The dataset utilized in this experiment is a collection of 1427 X-Ray images. 224 images with confirmed Covid-19, 700 images with confirmed common pneumonia, and 504 images of normal conditions are included. The data was collected from the available X-Ray images on public medical repositories. With transfer learning, an overall accuracy of 97.82% in the detection of Covid-19 is achieved.

• The paper demonstrates that CNN-based transfer learning achieves high diagnostic accuracy for COVID-19 detection, with VGG19 reaching 98.75% accuracy and MobileNet reducing false negatives.
• The study employs both feature extraction and fine-tuning strategies to adapt pre-trained CNN models for analyzing standardized thoracic X-ray images.
• The evaluation via 10-fold cross-validation highlights robust model performance, indicating potential for rapid, non-contact COVID-19 screening in clinical settings.

An Evaluation of CNN-Based Transfer Learning for Covid-19 Detection via X-Ray Imaging

This essay examines the effectiveness of using state-of-the-art Convolutional Neural Networks (CNNs) combined with transfer learning for the automatic detection of Covid-19 from thoracic X-ray images, as presented in the paper by Ioannis D. Apostolopoulos and Tzani Bessiana. The researchers utilized a dataset containing X-ray images to develop a model for accurately identifying Covid-19, common pneumonia, and normal conditions.

Dataset and Methodology

The dataset in this study comprises 1427 X-ray images, including 224 confirmed Covid-19 cases, 700 cases of common pneumonia, and 504 normal conditions. The data was sourced from public medical repositories, including the Radiological Society of North America (RSNA) and Radiopaedia. Each image was rescaled to a uniform size of 200x266 pixels to maintain consistency across the dataset.

The research emphasizes the application of transfer learning, a technique wherein pre-trained models on large datasets are fine-tuned on a smaller, task-specific dataset. This approach leverages the feature extraction capabilities already learned by the model in a broader context. The authors employed multiple CNN architectures, including VGG19, MobileNet, Inception, Xception, and Inception ResNet v2, to evaluate their performance in detecting Covid-19.

Transfer Learning with CNNs

The paper discusses two primary strategies for transfer learning:

1. Feature Extraction via Transfer Learning: This involves retaining the entire architecture and pre-trained weights of the model and using it as a feature extractor. The output features are then fed into a new network for classification.
2. Fine-Tuning: This involves making certain layers of the pre-trained model trainable while keeping the rest frozen. Adjustments are made to the architecture and parameters to optimize performance for the new dataset.

The CNNs were trained using specific hyperparameters such as the Rectified Linear Unit (ReLU) activation function, dropout layers to prevent overfitting, and the Adam optimization method. Training was conducted over ten epochs with a batch size of 64.

Results

The evaluation was performed using 10-fold cross-validation, and the performance metrics recorded include accuracy, sensitivity, and specificity. VGG19 and MobileNet emerged as the top-performing models. The critical findings are as follows:

• VGG19: Achieved an accuracy of 98.75% for Covid-19 detection (2-class) and 93.48% for multi-class classification (3-class). It recorded a sensitivity of 92.85% and a specificity of 98.75%.
• MobileNet: Achieved an accuracy of 97.40% for Covid-19 detection (2-class) and 92.85% for multi-class classification (3-class). It recorded a sensitivity of 99.10% and a specificity of 97.09%.

The confusion matrix revealed that MobileNet had a lower number of False Negatives compared to VGG19, making it a more effective model for the specific task of Covid-19 detection.

Discussion and Implications

The study demonstrates that transfer learning with CNNs can achieve high accuracy in the automatic detection of Covid-19 from X-ray images. Specifically, the MobileNet model’s ability to minimize False Negatives is crucial for the practical implementation of such diagnostic tools, as undetected cases can have significant public health implications.

Future research should address several limitations, including the need for a larger dataset, especially for Covid-19 cases. Additionally, models should be developed to distinguish Covid-19 from other similar viral infections like SARS and a wider variety of pneumonia cases.

This research contributes to the potential development of a low-cost, rapid, and automated diagnostic tool for Covid-19, which could significantly augment clinical decision-making without requiring direct contact with patients. This is particularly advantageous in mitigating the spread of the virus during pandemics.

Conclusion

The paper by Apostolopoulos and Bessiana provides a substantial contribution to the application of deep learning in medical imaging, specifically the use of CNN-based transfer learning for Covid-19 detection. The findings underscore the efficacy of pre-trained CNNs in achieving high diagnostic accuracy and pave the way for future improvements and broader clinical applications in medical image analysis.