A novel classification model based on cerebral 18F-FDG uptake pattern facilitates the diagnosis of acute/subacute seropositive autoimmune encephalitis

Abstract

PurposeTo explore the intrinsic alteration of cerebral 18F-FDG metabolism in acute/subacute seropositive autoimmune encephalitis (AE) and to propose a universal classification model based on 18F-FDG metabolic patterns to predict AE. MethodsCerebral 18F-FDG PET images of 42 acute/subacute seropositive AE patients and 45 healthy controls (HCs) were compared using voxelwise and region of interest (ROI)-based schemes. The mean standardized uptake value ratios (SUVRs) of 59 subregions according to a modified Automated Anatomical Labeling (AAL) atlas were compared using a t-test. Subjects were randomly divided into a training set (70%) and a testing set (30%). Logistic regression models were built based on the SUVRs and the models were evaluated by determining their predictive value in the training and testing sets. ResultsThe 18F-FDG uptake pattern in the AE group was characterized by increased SUVRs in the brainstem, cerebellum, basal ganglia, and temporal lobe, and decreased SUVRs in the occipital, and frontal regions with voxelwise analysis (false discovery rate [FDR] p<0.05). Utilizing ROI-based analysis, we identified 15 subareas that exhibited statistically significant changes in SUVRs among AE patients compared to HC (FDR p<0.05). Further, a logistic regression model incorporating SUVRs from the calcarine cortex, putamen, supramarginal gyrus, cerebelum_10, and hippocampus successfully enhanced the positive predictive value from 0.76 to 0.86 when compared to visual assessments. This model also demonstrated potent predictive ability, with AUC values of 0.94 and 0.91 observed for the training and testing sets, respectively. ConclusionsDuring the acute/subacute stages of seropositive AE, alterations in SUVRs appear to be concentrated within physiologically significant regions, ultimately defining the general cerebral metabolic pattern. By incorporating these key regions into a new classification model, we have improved the overall diagnostic efficiency of AE.