Researchers have developed a deep learning-based AI algorithm that automatically analyzes chest X-rays to rapidly detect COVID-19 infection with more than 98% accuracy, distinguishing between normal X-rays and X-rays from people with pneumonia, which often presents with the same symptoms as COVID.
Real-time reverse transcription-polymerase chain reaction (RT-PCR) testing is the most widely used method of diagnosing COVID-19 infection. But there are issues with using real-time PCR testing: it’s costly, results can be slow, and it’s prone to producing false negatives. As an adjunct, CT scans of the chest and chest X-rays also play a role in the timely detection and management of contagious infections, especially when RT-PCR returns a negative result.
COVID-19 produces particular radiological ‘signatures’ in chest X-rays that radiologists use to diagnose infection with the virus. However, meticulously examining X-rays for signs of infection is time-consuming and, because it relies on the human eye, may not always be accurate. So, researchers at the University of Technology Sydney (UTS) enlisted the help of AI to streamline the diagnostic process.
“The most widely used COVID-19 test, real-time polymerase chain reaction (PCR), can be slow and costly and produces false negatives,” said Amir Gandomi, corresponding author of the study. “To confirm a diagnosis, radiologists need to manually examine CT scans or X-rays, which can be time-consuming and prone to error.”
The other complicating factor is that the symptoms of COVID-19 infection – fever, cough, difficulty breathing, sore throat – can be difficult to distinguish from other respiratory viral infections, such as flu or pneumonia.
In recent years, machine learning algorithms have gained popularity in medicine, assisting doctors in diagnosing Parkinson’s disease, detecting breast cancer, and predicting stroke and heart failure. Deep learning, a subfield of AI, is particularly well-suited to creating a model that can produce accurate results from input data without requiring manual feature extraction. In the current study, the researchers developed a deep-learning-based algorithm called a Custom Convolutional Neural Network (Custom-CNN), specifically designed for diagnosing COVID-19.
Two freely available chest X-ray datasets were used to test and train the AI model. The datasets comprised three categories of chest X-ray images: normal, coronavirus-positive, and viral pneumonia. To train the Custom-CNN model, 80% of the total images were used, while 20% were reserved for testing.
The objective of the study was to assess the model's effectiveness in examining various relationships, including coronavirus and viral pneumonia, normal and viral pneumonia, and coronavirus and normal, and the associations among the three classes of X-ray images. Results demonstrated that the Custom-CNN model achieved a classification accuracy of 98.19% in its classification of COVID, normal and pneumonia image samples. Comparing the model’s results to those obtained using other models, the Custom-CNN outperformed them all.
“Deep learning offers an end-to-end solution, eliminating the need to manually search for biomarkers,” Gandomi said. “The Custom-CNN model streamlines the detection process, providing a faster and more accurate diagnosis of COVID-19.”
The early diagnosis of COVID-19 infection can ensure that patients get the correct treatment, including antivirals, which work best if taken within five days of the onset of symptoms. It could also encourage them to isolate and protect others from getting infected.
“The new AI system could be particularly beneficial in countries experiencing high levels of COVID-19 where there is a shortage of radiologists,” said Gandomi. “Chest X-rays are portable, widely available and provide lower exposure to ionizing radiation than CT scans.”
The study was published in the journal Scientific Reports.
Source: UTS
