TY - JOUR T1 - Examination of the relationship between PiB-PET and FDG-PET in Alzheimer’s disease using random forest and two-sample t-test JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 3105 LP - 3105 VL - 61 IS - supplement 1 AU - Yuma Tsubaki AU - Go Akamatsu AU - Natsumi Shimokawa AU - Aya Takashima AU - Suguru Katsube AU - Hideaki Sato AU - Kodai Kumamoto AU - Masayuki Sasaki Y1 - 2020/05/01 UR - http://jnm.snmjournals.org/content/61/supplement_1/3105.abstract N2 - 3105Purpose: Pittsburgh compound-B positron emission tomography (PiB-PET) quantitatively measures amyloid-β accumulation and fluorodeoxy-D-glucose PET (FDG-PET) measures glucose metabolism in the cerebral cortex. These measurements are used to diagnose and assess therapeutic effects in Alzheimer’s disease (AD) using quantitative indexes such as standardized uptake value ratio (SUVR) and t-values. In this study, the relationship between amyloid-β accumulation and glucose hypometabolism in early AD was investigated. Materials and Methods: A total of 164 subjects (46 AD subjects, 60 mild cognitive impairment (MCI) subjects, and 58 normal control (NC) subjects) who underwent PiB-PET and FDG-PET through the Japanese Alzheimer’s disease neuroimaging initiative were analyzed. Standard visual evaluation of PiB-PET was assessed as positive (+) or negative (−). Clinical data were provided by the National Bioscience Database Center. This study was approved by the ethics committee. SPM12, PMOD, and Python3.6 were used for analysis. AAL-ROI based on the anatomical brain region was used as the region of interest. SUVR and t-value in 62 regions were calculated. The t-value for each voxel was calculated by two-sample t-test between the NC and case group (AD and MCI). The results of SUVR were used to create a random forest classifier model. Feature importance, the quantitative value of the degree of effect on results for each explanatory variable, was calculated for each region. Results: In PiB-PET, the NC (+) group showed a significantly high t-value (p < 0.05, t > 5.66) for the cingulate gyrus, precuneus, and straight gyrus in the two-sample t-test. In the MCI (+) group and the AD (+) group, the t-values were significantly high for the whole brain (p < 0.05, t > 5.44). In FDG-PET, the cingulate gyrus, precuneus, hippocampus, and parahippocampal gyrus showed a high t-value (p < 0.05, t > 5.12) in the MCI (+) and the AD (+) groups. In SUVR analysis using machine learning, the calcarine sulcus, lingual gyrus, and angular gyrus were selected as the highest feature importance regions in PiB-PET. Accuracy for differentiation was 76% (NC vs MCI), 84% (NC vs AD), and 71% (MCI vs AD). In FDG-PET, the posterior cingulate gyrus, parahippocampal gyrus, and angular gyrus were selected as the highest feature importance region, and the accuracy for differentiation was 78% (NC vs MCI), 87% (NC vs AD), and 75% (MCI vs AD). Conclusions: In early AD, amyloid-β accumulation (measured using PiB-PET) and glucose hypometabolism (measured using FDG-PET) were commonly observed in the cingulate gyrus and precuneus. Both amyloid-β accumulation and glucose hypometabolism were commonly observed in some regions, although the region where the abnormal findings were first observed was different in PiB-PET and FDG-PET. ER -