Contributo da Inteligência Artificial para classificação de imagem em estudos de perfusão do miocárdio: uma revisão sistemática

Abstract

Introduction – Artificial Intelligence (AI) methods such as deep learning (DL) and, in particular, convolutional neural networks (CNN), are characterized by performing tasks that normally require human cognition. In myocardial perfusion studies, AI systems have assumed importance as an ancillary tool for nuclear medicine (NM) specialists, particularly for image classification in the context of coronary artery disease (CAD). Objectives – Evaluate DL methods' contribution to image classification in myocardial perfusion studies. Methods – A systematic review was carried out, which included 11 articles, searched in databases, PubMed, Web of Science, and Scopus. Studies published in the last five years that seek to evaluate the performance of DL methods for image classification in myocardial perfusion studies in the context of CAD were included. Results – Eleven articles were included in the systematic review, where 82% used CNN and the remaining 18% applied other DL methods. The Red-Green-Blue Convolutional Neural Network (RGB-CNN) architecture demonstrated greater applicability, corresponding to 45% of the CNN used, and presented better performance (AUC=94.58%). The remaining architectures corresponded to 11% each (CNN 2D; Inception V3; ResNet152V2; CNN Hand-Crafted and ResNet50). Conclusion – AI methods, namely DL using CNN, prove to be beneficial for the classification of myocardial perfusion images, with potential application in the early diagnosis of CAD, although with the need for further investigation.