Post-acute sequelae of SARS-CoV-2 infection (PASC) is characterized by persistent symptoms following SARS-CoV-2 infection. Children with PASC are at risk of developing cardiac complications. Echocardiography has been instrumental in identifying cardiac abnormalities. This study applies deep learning to enhance the detection and understanding of echocardiographic changes in children with PASC. A case-control study was conducted at a pediatric tertiary center in central Taiwan. Children under 18 years who tested positive for SARS-CoV-2 and experienced symptoms for longer than 4 weeks were recruited between July 1, 2022, and July 31, 2023, during the Omicron variant surge. Echocardiographic data were also collected from a control group, consisting of children who presented with similar symptoms and received medical care in the same pediatric tertiary center in 2018. Children with congenital or structural heart disease, inflammatory conditions, or arrhythmias were excluded. Echocardiographic images were analyzed using a ResNet-50-based deep learning model to identify cardiac abnormalities. A total of 270 children with PASC and 400 age-matched control children were included. Standard echocardiographic parameters, including EF, FS, chamber dimensions, and valvular assessment, did not reveal abnormalities in the PASC group. The deep learning model achieved an accuracy of 96.6%, sensitivity of 96.7%, specificity of 96.2%, and balanced accuracy of 96.4%. AI-assisted echocardiographic analysis demonstrated high performance in distinguishing cardiac function between PASC and controls. Deep learning models enhance the detection of subtle cardiac changes in children with PASC. Although the deep learning model demonstrated high performance in distinguishing PASC from controls, the clinical significance of these subtle image-based differences remains uncertain and requires further evaluation in large-scale studies with long-term follow-up.
Long Covid