Artificial Intelligence In Healthcare: Diagnosis Of Tuberculosis With Chest X-Ray Images But Without A Radiologist

Tuberculosis (TB) is an acute infectious disease that damages the lungs. It can be easily treated with the help of antibiotics, provided an early and accurate diagnosis is made. Diagnosis of tuberculosis can be done based on the X-ray images of the chest region. However, analyzing these X-ray films can only be done by an expert radiologist. Unfortunately, most places where tuberculosis is highly prevalent, viz. remote areas of developing and underdeveloped countries, lack skilled professional radiologists.

A team of researchers led by Baskaran Sundaram and Paras Lakhani from Thomas Jefferson University Hospital (TJUH), Philadelphia, has developed a highly innovative artificial intelligence model that can evaluate chest X-ray images to diagnose TB. The model can operate without the supervision of a radiologist, according to a EurekAlert public release.

Exemplary application of AI in healthcare artificial intelligence models is designed to accomplish complicated tasks without human interference. It functions by identifying discrepancies in the radiographs by comparing it to the database of already existing data. The researchers used this concept to train the computer systems to identify the changes in the radiographs of tuberculosis patients and healthy individuals.

According to Financial Express, Sundaram and his team used 1,007 X-ray images that included both TB and non-TB patients, for training the models. These images were obtained from the X-ray data sets of National Institutes of Health, TJUH and Belarus Tuberculosis Portal.

AlexNet and GoogLeNet were used for the development of the said artificial intelligence model. Once the training is completed, the models were used to screen 150 more X-ray images that were not included in the training.

Diagnosis was made based on the consensus decision of both the artificial intelligence models. Out of the 150 X-ray images tested, correct diagnosis was made in 137 cases (96 percent). A consensus could not be achieved in the remaining 13 images. Those cases were then recommended for expert analysis by a radiologist.

HealthDay reported the study findings that were published yesterday in the Radiology journal indicate that these models function with high accuracy. This makes them ideal for application in real time conditions. This novel application of artificial intelligence in healthcare can not only help in making fast and correct diagnosis of tuberculosis but also help in starting the treatment procedure at the earliest.