New AI Tool Can Detect COVID-19 From Chest X-Rays With Over 98% Accuracy

A team of researchers has unveiled a novel AI algorithm designed to swiftly and accurately detect COVID-19 infection from chest X-rays, achieving a remarkable accuracy rate exceeding 98%.

This AI system addresses the urgent need for efficient automated tools in detecting COVID-19, considering its profound impact on public health and the global economy.

"The new AI system could be particularly beneficial in countries experiencing high levels of COVID-19 where there is a shortage of radiologists. Chest X-rays are portable, widely available and provide lower exposure to ionizing radiation than CT scans," Professor Amir H. Gandomi, the corresponding author from the University of Technology Sydney (UTS) Data Science Institute, said in a statement.

AI Detects COVID-19

Gandomi emphasized the necessity of having effective diagnostic tools for COVID-19, particularly given the limitations and potential drawbacks of widely used testing methods like real-time polymerase chain reaction (PCR).

The PCR test, while commonly used, can be slow, expensive, and occasionally produce false-negative results, according to Gandomi. Additionally, confirming a diagnosis often requires manual examination of CT scans or X-rays by radiologists, a process prone to error and time-consuming.

Gandomi highlighted the significance of the newly developed AI system, especially in regions with a shortage of radiologists and high levels of COVID-19.

The AI system utilizes a sophisticated deep learning-based algorithm called a Custom Convolutional Neural Network (Custom-CNN). This algorithm is adept at swiftly and accurately distinguishing between COVID-19 cases, normal cases, and pneumonia in X-ray images, offering a comprehensive solution that streamlines the detection process without needing manual biomarker searches.

According to the researchers, the portability and widespread availability of chest X-rays, coupled with their lower exposure to ionizing radiation compared to CT scans, make them a valuable diagnostic tool, particularly in resource-constrained environments.

The Custom-CNN Model

Professor Gandomi highlighted the challenges in differentiating COVID-19 from other respiratory conditions, such as influenza and pneumonia, making accurate and efficient diagnostic tools crucial.

Through deep learning, the Custom-CNN model provides an end-to-end solution, aiding in faster and more precise COVID-19 diagnoses.

The AI system proves beneficial when PCR or rapid antigen tests yield negative or inconclusive results, requiring further radiological examination. While emphasizing the crucial role of radiologists in medical diagnosis, Professor Gandomi asserted that AI technology can complement their efforts, facilitating accurate and efficient diagnoses.

The performance of the Custom-CNN model was subjected to a comprehensive comparative analysis, focusing on accuracy as the primary performance criterion. The results indicated that the new model outperforms other AI diagnostic models, making it a promising tool for rapidly and accurately diagnosing COVID-19.

The findings of the research team were published in the journal Scientific Reports.