Abstract
AbstractBackgroundIn the analysis of brain fluorodeoxyglucose positron emission tomography (FDG-PET), intensity normalization is a necessary step to reduce inter-subject variability. However, the choice of an appropriate normalization method in stroke studies remains unclear, leading to inconsistent findings in the literature.Materials and methodsHere, we propose a regression-based and single-subject-based model for analyzing FDG-PET images without intensity normalization. Two independent data sets were collected before and after middle cerebral artery occlusion (MCAO), with one comprising 120 rats and the other comprising 96 rats. After data preprocessing, voxel intensities in the same region and hemisphere were paired before and after the MCAO scan. A linear regression model was applied to the paired data, and the coefficient of determination was calculated to measure the linearity. The results between the ipsilateral and contralateral hemispheres were compared, and significant regions were defined as those having reduced linearity. Our method was compared with existing intensity normalization methods and validated using the triphenyl tetrazolium chloride (TTC) staining data.ResultsThe proposed method detected more injury areas compared to existing approaches, as confirmed by the ground truth provided by TTC. The area under the curve (AUC) of the average receiver operating characteristic (ROC) curves using our method reached 0.84, whereas the AUCs using existing methods ranged from 0.77 ∼ 0.79. The average false positive rate (FPR) and true positive rate (TPR) of the individual analysis results using our method (FPR = 0.06, TPR = 0.56) were better than the group-wise analysis results usingt-tests (FPR = 0.10, TPR = 0.51). The identified injury regions were consistent in the two independent data sets. Some of them were confirmed by other publications.ConclusionsThe proposed method offers a new quantitative approach to analyzing FDG-PET images. The calculation does not involve intensity normalization and can be applied to a single subject. The method yields more sensitive results than existing intensity normalization methods.
Publisher
Cold Spring Harbor Laboratory