Deep Learning for Screening COVID-19 using Chest X-Ray Images (2004.10507v4)

Abstract: With the ever increasing demand for screening millions of prospective "novel coronavirus" or COVID-19 cases, and due to the emergence of high false negatives in the commonly used PCR tests, the necessity for probing an alternative simple screening mechanism of COVID-19 using radiological images (like chest X-Rays) assumes importance. In this scenario, ML and deep learning (DL) offer fast, automated, effective strategies to detect abnormalities and extract key features of the altered lung parenchyma, which may be related to specific signatures of the COVID-19 virus. However, the available COVID-19 datasets are inadequate to train deep neural networks. Therefore, we propose a new concept called domain extension transfer learning (DETL). We employ DETL, with pre-trained deep convolutional neural network, on a related large chest X-Ray dataset that is tuned for classifying between four classes \textit{viz.} $normal$, $pneumonia$, $other_disease$, and $Covid-19$. A 5-fold cross validation is performed to estimate the feasibility of using chest X-Rays to diagnose COVID-19. The initial results show promise, with the possibility of replication on bigger and more diverse data sets. The overall accuracy was measured as $90.13\% \pm 0.14$. In order to get an idea about the COVID-19 detection transparency, we employed the concept of Gradient Class Activation Map (Grad-CAM) for detecting the regions where the model paid more attention during the classification. This was found to strongly correlate with clinical findings, as validated by experts.

Deep Learning for Screening COVID-19 using Chest X-Ray Images: A Critical Overview

The paper "Deep Learning for Screening COVID-19 using Chest X-Ray Images," authored by Basu et al., explores the application of deep learning techniques to aid the screening of COVID-19 through thoracic imaging. The authors propose the use of chest X-ray images as a viable alternative to the conventional PCR tests, which are known for their complexity and high occurrence of false negatives.

Methodological Innovation

One notable contribution in this paper is the introduction of Domain Extension Transfer Learning (DETL). This approach is employed to handle the inadequacy of direct COVID-19 datasets for training purposes. DETL leverages pre-trained deep convolutional neural networks, tailored to recognize COVID-19 signatures in chest X-ray images. By training on an extended dataset incorporating multiple related disease classifications, DETL aims to enhance the diagnostic accuracy of the system.

Technical Implementation

The implementation utilizes transfer learning from large existing datasets, listing CNN architectures like AlexNet, VGGNet, and ResNet. This research emphasizes augmenting existing chest X-ray datasets with COVID-19 images to classify outcomes into four categories: normal, pneumonia, other diseases, and COVID-19. The authors detail the process of fine-tuning CNNs to adapt to a classification task specifically tailored for the four identified classes, using a systematic approach that reconfigures the last layers of the pre-existing networks.

Empirical Results

A 5-fold cross-validation was employed for model evaluation, demonstrating impressive accuracy rates: VGGNet achieved the highest accuracy at approximately 90.13%, outperforming AlexNet and ResNet. These results suggest the efficacy of the proposed DETL framework in providing rapid and reliable COVID-19 identification through chest X-ray analysis, notwithstanding some misclassification between pneumonia and other diseases due to co-occurrence.

Decision Visualization

To enhance interpretability, the authors apply Grad-CAM, a technique that aids in highlighting the regions of interest within the images that the model focuses on when making predictions. This visualization aligns with clinical observations of features such as ground-glass opacities in COVID-19 patients, thereby reinforcing the clinical applicability of the deep learning-based screening framework.

Implications and Future Work

This paper presents significant implications for the deployment of AI-assisted diagnostics in global healthcare systems, potentially easing the burden on facilities overwhelmed by COVID-19 cases. By offering an alternative screening method, the research has practical relevance for quick and widespread testing. However, the outcomes advocate for expanding datasets to improve the robustness and generalizability of the models further.

Future work may concentrate on broadening dataset inclusivity and refining the DETL methodology to bolster predictive accuracy across diverse populations. Exploring additional imaging modalities and integrating multi-modal data could offer comprehensive diagnostic tools in combating not only COVID-19 but other emergent respiratory illnesses.

Overall, Basu et al. contribute to the multidisciplinary dialogue on AI applications in healthcare, underpinning the need for continual advancements in computational methods tailored to evolving medical challenges.