Deep-COVID: Predicting COVID-19 From Chest X-Ray Images Using Deep Transfer Learning (2004.09363v3)

Abstract: The COVID-19 pandemic is causing a major outbreak in more than 150 countries around the world, having a severe impact on the health and life of many people globally. One of the crucial step in fighting COVID-19 is the ability to detect the infected patients early enough, and put them under special care. Detecting this disease from radiography and radiology images is perhaps one of the fastest ways to diagnose the patients. Some of the early studies showed specific abnormalities in the chest radiograms of patients infected with COVID-19. Inspired by earlier works, we study the application of deep learning models to detect COVID-19 patients from their chest radiography images. We first prepare a dataset of 5,000 Chest X-rays from the publicly available datasets. Images exhibiting COVID-19 disease presence were identified by board-certified radiologist. Transfer learning on a subset of 2,000 radiograms was used to train four popular convolutional neural networks, including ResNet18, ResNet50, SqueezeNet, and DenseNet-121, to identify COVID-19 disease in the analyzed chest X-ray images. We evaluated these models on the remaining 3,000 images, and most of these networks achieved a sensitivity rate of 98% ($\pm$ 3%), while having a specificity rate of around 90%. Besides sensitivity and specificity rates, we also present the receiver operating characteristic (ROC) curve, precision-recall curve, average prediction, and confusion matrix of each model. We also used a technique to generate heatmaps of lung regions potentially infected by COVID-19 and show that the generated heatmaps contain most of the infected areas annotated by our board certified radiologist. While the achieved performance is very encouraging, further analysis is required on a larger set of COVID-19 images, to have a more reliable estimation of accuracy rates. The dataset, model implementations (in PyTorch), and evaluations, are all made publicly available for research community at https://github.com/shervinmin/DeepCovid.git

Deep-COVID: Predicting COVID-19 From Chest X-Ray Images Using Deep Transfer Learning

The paper "Deep-COVID: Predicting COVID-19 From Chest X-Ray Images Using Deep Transfer Learning" presents a deep learning-based framework tailored for the detection of COVID-19 from chest X-ray images. The authors, led by Shervin Minaee, compiled and utilized a dataset of 5,000 chest X-ray images, referred to as COVID-Xray-5k, to fine-tune four popular convolutional neural network (CNN) models: ResNet18, ResNet50, SqueezeNet, and DenseNet-121, by leveraging transfer learning techniques.

Methodology

The deep learning framework involves fine-tuning pre-trained CNN models on the COVID-Xray-5k dataset, which consists of chest X-ray images from publicly available datasets. Specifically, the dataset contains 2,084 training images and 3,100 test images. COVID-19 positive cases were identified by board-certified radiologists, ensuring the reliability of the labels. Two main strategies were adopted to address the limited number of COVID-19 images:

1. Data augmentation to increase the number of samples by a factor of 5.
2. Fine-tuning the last layer of the pre-trained models instead of training from scratch.

Models and Training

The fine-tuned models, namely ResNet18, ResNet50, SqueezeNet, and DenseNet-121, underwent training with a cross-entropy loss function optimized using the ADAM optimizer. The images were resized to 224x224 pixels before being fed into the neural networks, consistent with the input requirements of the pre-trained models.

Experimental Results

The evaluation metrics used include sensitivity, specificity, ROC curve, and precision-recall curve. On the test set of 3,100 images, the top-performing model achieved a sensitivity rate of 98% with a specificity rate of approximately 90.7%. ResNet18 and SqueezeNet demonstrated slightly superior performance compared to ResNet50 and DenseNet-121. More specifically, SqueezeNet achieved a sensitivity rate of 98% and a specificity rate of 92.9%.

The results indicate that the models are proficient in distinguishing COVID-19 positive images from non-COVID images, emphasizing the capability of deep learning frameworks in medical image analysis.

Implications and Future Directions

The findings of this paper are promising for the application of deep learning models in the rapid identification of COVID-19 positive cases, which can significantly aid in early diagnosis and isolation to control the spread of the virus. The publicly available COVID-Xray-5k dataset can serve as a benchmark for future research in this domain.

However, the authors pointed out that the small number of available COVID-19 images limits the reliability of the sensitivity and specificity rates. Future work should focus on expanding the dataset to include more COVID-19 positive images to improve the reliability of the model evaluations. Additionally, further analysis on a larger and more diverse set of labeled images is recommended to refine the model accuracy and robustness.

Conclusion

This paper contributes significantly to the field of medical image analysis by demonstrating the efficacy of deep learning models for COVID-19 detection from chest X-ray images. The authors provide an extensive evaluation of multiple CNN architectures, which have shown promising results while addressing the challenges related to the scarcity of labeled COVID-19 data. Future endeavors should focus on enlarging the dataset and validating the models in various clinical settings to further establish the utility of these AI-based diagnostic tools in real-world scenarios.