Value of dynamic metabolic curves and artificial neural network prediction models based on F-18-FDG PET/CT multiphase imaging in differentiating nonspecific solitary pulmonary lesions: a pilot study

ObjectiveTo investigate the value of dynamic metabolic curves and artificial neural network prediction models based on F-18-FDG PET multiphase imaging in differentiating nonspecific solitary pulmonary lesions. MethodsThis study enrolled 71 patients with solitary pulmonary lesions (48 malignant and 23 benign lesions) who underwent multiphase F-18-fluorodeoxyglucose (F-18-FDG)-PET/CT imaging. We recorded information on age, sex and uniformity of FDG uptake, measured standardized uptake value, metabolic tumor volume and total lesion glycolysis at various time points, and calculated individual standardized uptake values, retention index (RI) and slope of metabolic curve. Variables with high diagnostic efficiency were selected to fit dynamic metabolic curves for various lesions and establish different artificial neural network prediction models. ResultsThere were no significant differences in the retention index, metabolic tumor volume, total lesion glycolysis or sex between benign and malignant lesions; standardized uptake values, the slopes of five metabolic curves, uniformity of FDG uptake, and age showed significant differences. Dynamic metabolic curves for various solitary pulmonary lesions exhibited characteristic findings. Model-1 was established using metabolic parameters with high diagnostic efficacy (area under the curve, 83.3%). Model-2 was constructed as Model-1 + age (area under the curve, 86.7%), whereas Model-3 was established by optimizing Model-2 (area under the curve, 86.0%). ConclusionsDynamic metabolic curves showed varying characteristics for different lesions. Referring to these findings in clinical work may facilitate the differential diagnosis of nonspecific solitary pulmonary lesions. Establishing an artificial neural network prediction model would further improve diagnostic efficiency.