RADIOLOGY AND AI: MODERN ALGORITHMS FOR DETECTING COVID-19 AND PNEUMONIA FROM LUNG IMAGES

• Mukaddas Fazlitdinova

  Tashkent State Medical University No. 1 of the General Medicine Faculty, Student Group 109 Tashkent, Uzbekistan

  Author

• Fazliddin Arzikulov

  Assistant of the Department of Biomedical Engineering, Informatics and Biophysics at the Tashkent State Medical University

  Author

Coronavirus disease is a global pandemic that, to date, has infected millions of people worldwide and caused the deaths of thousands. Any technological tool that allows for rapid and highly accurate detection of COVID-19 is extremely important for healthcare professionals. Currently, the main clinical method for diagnosing COVID-19 is laboratory testing based on reverse transcriptase and polymerase chain reaction (RT-PCR). However, this method is expensive, has relatively low sensitivity, and requires specially trained medical personnel. Therefore, easily accessible and rapid imaging methods, such as radiography, can serve as an effective alternative for detecting COVID-19.This study was conducted to evaluate the effectiveness of artificial intelligence technologies in the rapid and accurate detection of COVID-19 from chest X-ray images. The primary objective of the research was to develop a reliable and high-accuracy method for the automatic detection of COVID-19–related pneumonia in digital chest X-ray images using pre-trained deep learning–based algorithms. The authors created a comprehensive open dataset by combining several publicly available databases and collecting images from recently published scientific articles. The dataset included chest X-ray images of 423 COVID-19 patients, 1,485 cases of viral pneumonia, and 1,579 healthy individuals.To train and evaluate the pre-trained deep learning models, image augmentation techniques were applied along with transfer learning. The models were trained and tested according to two classification schemes: the first scheme differentiated between healthy individuals and COVID-19–related pneumonia, while the second scheme classified healthy individuals, viral pneumonia, and COVID-19–related pneumonia with and without image augmentation.For the first classification scheme, accuracy, precision, sensitivity, and specificity were 99.7%, 99.7%, 99.7%, and 99.55%, respectively. For the second scheme, these metrics were 97.9%, 97.95%, 97.9%, and 98.8%, respectively. The high accuracy of this computer-aided diagnostic system can significantly enhance the speed and reliability of COVID-19 detection. This is particularly important in the context of a pandemic, given the high disease burden and limited availability of medical resources.

• pdf

This work is licensed under a Creative Commons Attribution 4.0 International License.