Extracting possibly representative COVID-19 Biomarkers from X-Ray images with Deep Learning approach and image data related to Pulmonary Diseases

Abstract: In this study, the problem of automatically classifying pulmonary diseases, including the recently emerged COVID-19, from X-Ray images, is considered. While the spread of COVID-19 is increased, new, automatic, and reliable methods for accurate detection are essential to reduce the exposure of the medical experts to the outbreak. X-ray imaging, although limited to specific visualizations, may be helpful for the diagnosis. Deep Learning has proven to be a remarkable method to extract massive high-dimensional features from medical images. Specifically, in this paper, the state-of-the-art Convolutional Neural Network called Mobile Net is employed and trained from scratch to investigate the importance of the extracted features for the classification task. A large-scale dataset of 3905 X-Ray images, corresponding to 6 diseases is utilized for training MobileNet v2, which has been proven to achieve remarkable results in related tasks. The results suggest that training CNNs from scratch may reveal vital biomarkers related but not limited to the COVID-19 disease, while an overall classification accuracy of the seven classes reaches 87.66%. Besides, this method achieves 99.18% accuracy, 97.36% Sensitivity, and 99.42% Specificity in the detection of COVID-19.

• The paper demonstrates that training MobileNet v2 from scratch yields 99.18% accuracy in detecting COVID-19 from X-ray images.
• It compares three deep learning strategies, revealing that customized training on specific COVID-19 data outperforms transfer learning and fine-tuning.
• The study highlights the clinical potential of CNN-derived biomarkers for enhancing early diagnosis and reducing healthcare exposure risks.

Extracting COVID-19 Biomarkers from X-Ray Images with Deep Learning

The paper "Extracting Possibly Representative COVID-19 Biomarkers from X-Ray Images with Deep Learning" addresses the pressing need for effective and efficient methods for the diagnosis of COVID-19 using medical imaging technologies, particularly amid the challenges posed by the global pandemic. The study demonstrates the application of a state-of-the-art Convolutional Neural Network (CNN), specifically MobileNet v2, for the automatic classification of X-ray images to identify pulmonary diseases including COVID-19.

Methodology

The research leverages a sizeable dataset comprising 3905 X-ray images, with 455 images of confirmed COVID-19 cases, alongside images of other pulmonary diseases such as Pulmonary Edema, Pleural Effusion, Chronic Obstructive Pulmonary Disease (COPD), and Pulmonary Fibrosis. The dataset construction involved curating X-ray images from several credible sources, including repositories from medical societies and research initiatives.

The authors investigate three distinct deep learning strategies for feature extraction: (a) training the CNN from scratch, (b) feature extraction using transfer learning, and (c) a hybrid approach known as fine-tuning. Each approach aims to assess the ability of MobileNet v2 to discern characteristic features—potential biomarkers—within the chest X-ray images.

Results and Findings

The study highlights significant findings through comprehensive experimentation. Training MobileNet v2 from scratch yielded remarkable classification performance, achieving 87.66% accuracy across seven classes of pulmonary conditions and an impressive 99.18% accuracy in detecting COVID-19 specifically. The sensitivity and specificity for COVID-19 detection stood at 97.36% and 99.42%, respectively, which underscores the reliability of the model in identifying true positive COVID-19 cases while minimizing false positives.

Conversely, transfer learning with of-the-self features and fine-tuning approaches underperformed compared to training from scratch, suggesting that newly discovered features—possibly unique to COVID-19—are more effectively extracted through direct learning from the specific data. These results support the hypothesis that some high-dimensional features mined via deep learning may function as reliable biomarkers indicative of COVID-19.

Implications and Future Work

This work underscores the potential of CNNs trained from scratch to identify informative features that existing models may overlook. The findings advocate for further investigation into the biological significance of these features, considering the interpretability challenges inherent in deep learning applications. Future studies should aim to validate these biomarker-like features in larger, more diverse datasets, including mild COVID-19 cases, to refine the diagnostic capabilities of AI models.

Furthermore, the paper suggests extending the feature extraction methods to identify more pulmonary diseases, thereby enhancing pre-diagnosis capabilities through automated systems. Such advancements could substantially reduce the exposure risk of healthcare providers to the virus and expedite the application of quarantine measures.

In conclusion, this paper contributes to the ongoing efforts in medical AI by reinforcing the capability of deep learning techniques to discern distinctive biomarkers from medical images, offering a robust alternative to traditional diagnostic methods amidst the COVID-19 crisis. Such innovation could pave the way for more responsive, scalable, and accessible diagnostic solutions in the future.