A Seoul National University Bundang Hospital (SNUBH) research team has released its research method based on a deep learning algorithm to diagnose sinusitis accurately.
|From left, Professors Sun Woo-joon and Lee Kyung-joon|
The most common way to diagnose sinusitis -- common inflammation of the paranasal sinuses, the cavities that produce the mucus necessary for the nasal passages to work effectively - is X-ray. Although the process is advantageous as it emits less radiation compared to computed tomography (CT), the method has a low accuracy than a CT scan. Therefore, physicians still use CT in making a precise diagnosis for surgeries.
The team, led by Professors Sun Woo-joon and Lee Kyung-joon of the department of radiology at the hospital, suggested that deep learning algorithms could improve the diagnostic accuracy of X-rays in diagnosing sinusitis.
To confirm their theory, the researchers organized 9,000 suspected cases as either normal or having sinusitis according to the X-ray findings, and divided the data into training and verification data to develop a deep learning algorithm. They also prepared two test datasets diagnosed according to the results of a CT examination and compared the diagnostic accuracy with five radiologists to verify the algorithm more precisely.
The performance of the deep-running algorithm showed the same level of diagnostic accuracy as the radiologist in all the experimental datasets. The team also confirmed that the accuracy was consistent when applied to data from Seoul National University Hospital.
"The study demonstrated that by using the deep learning algorithm, we could accurately diagnose sinusitis with a simple X-ray image," Professor Sun said. "Also, the use of X-rays can contribute to minimizing the patient's radiation exposure as the radiation dose compared with CT examination is only about one-twentieth."
Investigative Radiology published the result of the study.
<© Korea Biomedical Review, All rights reserved.>