Modelo computacional baseado em inteligência artificial para leitura dos padrões pulmonares na tomografia computadorizada da COVID-19 pós-aguda

Abstract

The pandemic brought catastrophic effects to public health, but it enabled the development of new knowledge in several areas of science, such as the implementation of artificial intelligence (AI). After three years, there are patients without complete remission of the disease or postacute COVID-19. Artificial intelligence (AI) may be relevant in detecting pulmonary complications and defining tomographic patterns associated with respiratory complications. This research focuses on two critical aspects of COVID-19: the pulmonary complications that can occur during the post-acute phase of the disease and the use of artificial intelligence (AI) to diagnose and predict complications. It aims to develop AI models for the recognition of Postacute COVID-19 chest tomographic (CT) patterns. To achieve this, radiologists analyzed 87 CT scans to establish CT patterns, train and test deep learning models. The best model was then selected for reading 8 complete CT scans. The results of the chosen model showed an average accuracy of 91,14% in detecting post-COVID patterns. Although the sample size was limited, the successful development of AI models for reading post-COVID-19 patterns on CT despite the small sample size. The sample used in the study reflects the epidemiological profile found in the literature. With the coherent and satisfactory results in the evaluation of the best model with the 8 complete CT scans, comparing the reading with the report of a radiologist. However, a larger sample size would be beneficial for future research