The modified deep learning architecture can effectively distinguish between COVID-19 and viral pneumonia.
The COVID-19 pandemic has highlighted the importance of early detection and treatment to save lives. Currently, the two main imaging tools used for determining COVID-19 infection are chest X-ray and computerised tomography (CT) of the chest. However, since COVID-19 cases are almost comparable to the different types of viral pneumonia infections, identifying COVID-19 is made more challenging as many of COVID-19’s features overlap with the other inflammatory diseases of the lung.
Today, many biomedical problems, such as cancer, apply artificial intelligence and deep learning techniques to diagnose the issues. The image features, which are immediately apparent in the original images, can be revealed using deep learning methods.
In a new study published in Biomedical Engineering: Applications, Basis and Communications, researchers at the Indian Institute of Technology developed a modified deep learning architecture to classify various chest X-ray images to detect COVID-19 and viral pneumonia.
When compared to other current approaches, the classification metrics showed a significant improvement. The team’s proposed classifier is able to independently extract feature maps and learn about the indices of these feature maps, which is used for the reconstruction of the classes. As such, it efficiently learns the variations of texture for different classes and is able to differentiate the two nearly related classes—COVID-19 and viral pneumonia.
Given the time sensitivity of COVID-19, the work will be helpful for medical diagnosis and allow for the efficient diagnosis of COVID-19 in the early stage. The proposed method can also be helpful for finding patterns of various related diseases that share common features. With an increased dataset, the effectiveness of the classification process can be further improved. [APBN]
Source: Tripathi, S., & Sharma, N. (2023). AUTOMATIC DETECTION OF COVID-19 AND VIRAL PNEUMONIA IN X-RAY IMAGES USING DEEP LEARNING APPROACH. Biomedical Engineering: Applications, Basis and Communications, 2350001.