Can AI help in screening Viral and COVID-19 pneumonia? (2003.13145v3)

Abstract: Coronavirus disease (COVID-19) is a pandemic disease, which has already caused thousands of causalities and infected several millions of people worldwide. Any technological tool enabling rapid screening of the COVID-19 infection with high accuracy can be crucially helpful to healthcare professionals. The main clinical tool currently in use for the diagnosis of COVID-19 is the Reverse transcription polymerase chain reaction (RT-PCR), which is expensive, less-sensitive and requires specialized medical personnel. X-ray imaging is an easily accessible tool that can be an excellent alternative in the COVID-19 diagnosis. This research was taken to investigate the utility of AI in the rapid and accurate detection of COVID-19 from chest X-ray images. The aim of this paper is to propose a robust technique for automatic detection of COVID-19 pneumonia from digital chest X-ray images applying pre-trained deep-learning algorithms while maximizing the detection accuracy. A public database was created by the authors combining several public databases and also by collecting images from recently published articles. The database contains a mixture of 423 COVID-19, 1485 viral pneumonia, and 1579 normal chest X-ray images. Transfer learning technique was used with the help of image augmentation to train and validate several pre-trained deep Convolutional Neural Networks (CNNs). The networks were trained to classify two different schemes: i) normal and COVID-19 pneumonia; ii) normal, viral and COVID-19 pneumonia with and without image augmentation. The classification accuracy, precision, sensitivity, and specificity for both the schemes were 99.7%, 99.7%, 99.7% and 99.55% and 97.9%, 97.95%, 97.9%, and 98.8%, respectively.

An Overview of Deep Learning in COVID-19 Pneumonia Detection Using Chest X-ray Images

The research paper titled "Can AI help in screening Viral and COVID-19 pneumonia?" explores an advanced methodological approach using convolutional neural networks (CNNs) for the classification of COVID-19 pneumonia using chest X-ray images. This investigation addresses a pivotal need for efficient and rapid diagnosis tools amidst the global COVID-19 pandemic, which has tested the limits of healthcare systems worldwide.

Methodology

The paper utilized deep learning methodologies, particularly pre-trained CNN models, to detect COVID-19 from chest X-ray (CXR) images. The researchers prepared a comprehensive public database composed of 423 COVID-19, 1485 viral pneumonia, and 1579 normal CXR images sourced from public databases and publications.

Data Augmentation and Transfer Learning

Two experimental setups were employed:

1. Binary Classification: Differentiating between COVID-19 pneumonia and normal cases.
2. Ternary Classification: Distinguishing among normal, viral pneumonia, and COVID-19 pneumonia cases.

For both setups, CNNs were trained and validated using transfer learning techniques applied to pre-trained networks such as MobileNetv2, SqueezeNet, ResNet18, ResNet101, DenseNet201, VGG19, InceptionV3, and CheXNet. The models were evaluated with and without image augmentation techniques to enhance the diversity and robustness of the training set. The stochastic gradient descent (SGD) optimizer facilitated efficient model training. Performance was validated using 5-fold cross-validation, ensuring robustness and generalizability.

Results

The paper details the exceptional performance metrics achieved by the CNN models:

• Binary Classification: With image augmentation, models achieved an impressive classification accuracy of up to 99.7%, and precision, recall, and specificity values exceeded 99%.
• Ternary Classification: Models recorded an accuracy of up to 97.94% with image augmentation, with significant values for precision, recall, and specificity.

DenseNet201 emerged as the most effective model, demonstrating superior performance in both binary and ternary classifications when trained on augmented data, outperforming even the well-regarded CheXNet.

Discussion

The use of deep CNNs, particularly within the transfer learning paradigm, demonstrates a promising pathway for the rapid and accurate screening of COVID-19 pneumonia. The use of image augmentation significantly enhanced the model performance, indicating the critical role of diversified image data in training robust neural networks.

The insight gained from activation maps suggests that the deeper convolutional layers can capture distinct features that are diagnostic of COVID-19, features that may not be readily apparent to human radiologists.

Implications and Future Directions

This research highlights the utility of AI and deep learning in assisting healthcare professionals by providing reliable, fast, and highly accurate diagnostic tools, especially crucial during a pandemic where healthcare resources are strained. The high sensitivity and specificity demonstrated by the trained CNNs could potentially lead to reduced misdiagnosis rates, particularly in differentiating COVID-19 pneumonia from other types of viral pneumonia and normal cases.

Future research could focus on expanding the database further and leveraging unsupervised learning techniques to discover new, potentially more subtle features indicative of COVID-19. Additionally, extending this methodology to other radiological modalities, such as CT scans, could provide a more comprehensive diagnostic toolkit. The integration of AI diagnostic tools within clinical workflows could streamline patient management and resource allocation efforts during ongoing and future health crises.

Overall, the paper underscores the potential of AI-driven diagnostic tools in transforming healthcare delivery, ensuring timely and accurate disease detection, and addressing the emergent needs of a global pandemic landscape.