摘要： 目的 基于¹⁸F-FDG PET探讨抗谷氨酸脱羧酶65（GAD65）抗体相关脑炎的代谢模式，并分析其与临床和神经影像学特征的相关性。方法 纳入2018年1月至2024年7月首都医科大学附属北京天坛医院收治的25例抗GAD65抗体相关脑炎患者，均行¹⁸F-FDG PET显像，将每例患者全脑分割为116个脑区，计算每个脑区的小脑标准化摄取值比值（SUVRc）；采用主成分分析对SUVRc降维，k均值聚类算法将抗GAD65抗体相关脑炎患者分为2个集群，Pearson相关分析和偏相关分析探讨关键脑区代谢活性与临床和MRI体积的相关性。结果 主成分分析显示，主成分1贡献最大的5个脑区均位于枕叶，依次为左侧楔叶、右侧楔叶、右侧舌回、左侧枕下回和右侧枕下回。k均值聚类算法将抗GAD65抗体相关脑炎患者分为2个集群（集群1为14例、集群2为11例）。抗GAD65抗体相关脑炎患者2个集群与健康对照者左侧楔叶（F=7.946，P=0.000）、右侧楔叶（F=8.406，P=0.000）、右侧舌回（F=9.447，P=0.000）、左侧枕下回（F=17.036，P=0.000）和右侧枕下回（F=18.312，P=0.000）SUVRc均有统计学意义，进一步两两比较，与对照组相比，集群2左侧楔叶（t=3.809，P=0.000）、右侧楔叶（t=2.959，P=0.000）、右侧舌回（t=3.726，P=0.000）、左侧枕下回（t=3.948，P=0.000）和右侧枕下回（t=3.846，P=0.000）SUVRc增加，称为枕叶高代谢集群；集群1左侧楔叶、右侧楔叶、右侧舌回、左侧枕下回和右侧枕下回代谢活性有降低趋势，称为非枕叶高代谢集群。MRI体积分析显示，非枕叶高代谢集群与枕叶高代谢集群差异脑区为楔叶（t=- 2.214，P=0.038）、枕上回（t=- 2.213，P=0.038）、眶回（t=- 2.829，P=0.010）、颞上回外侧部（t=- 2.246，P=0.035）、枕颞内侧沟（侧副沟）和舌回沟（t=- 2.160，P=0.042）。结论 通过¹⁸F-FDG PET结合聚类分析识别出抗GAD65抗体相关脑炎的不同代谢模式，并揭示其与临床和神经影像学特征的关联性，加深对其病理生理学机制的理解。

关键词: 脑炎, 自身免疫疾病, 谷氨酸脱羧酶, 正电子发射断层显像术, 聚类分析

Abstract: Objective To investigate the metabolic patterns of anti-glutamate decarboxylase 65 (GAD65) antibody-associated encephalitis using ¹⁸F-FDG PET, and to analyze the correlations between these patterns and clinical or neuroimaging features. Methods A total of 25 patients with anti-GAD65 antibody-associated encephalitis admitted to Beijing Tiantan Hospital, Capital Medical University, from January 2018 to July 2024 were enrolled. All patients underwent ¹⁸F-FDG PET imaging. The whole brain of each patient was segmented into 116 brain regions, and cerebellum-normalized standard uptake value ratio (SUVRc) was calculated for each region. Principal component analysis (PCA) was used to reduce the dimensionality of SUVRc, and k-means clustering algorithm was applied to divide the patients into 2 clusters. Pearson and partial correlation analyses were performed to explore the correlations between metabolic activity of key brain regions and clinical indicators or MRI volumes. Results PCA showed that the top 5 brain regions contributing most to Principal Component 1 were all located in the occipital lobe, in order of left cuneus, right cuneus, right lingual gyrus, left inferior occipital gyrus and right inferior occipital gyrus. The k-means clustering algorithm divided the patients into 2 clusters (14 cases in Cluster 1, 11 cases in Cluster 2). There were statistically significant differences in SUVRc of the left cuneus (F = 7.946, P = 0.000), right cuneus (F = 8.406, P = 0.000), right lingual gyrus (F = 9.447, P = 0.000), left inferior occipital gyrus (F = 17.036, P = 0.000), and right inferior occipital gyrus (F = 18.312, P = 0.000) between the 2 clusters and healthy controls. Further pairwise comparisons revealed that compared with the health controls, Cluster 2 had increased SUVRc in the left cuneus (t = 3.809, P = 0.000), right cuneus (t = 2.959, P = 0.000), right lingual gyrus (t = 3.726, P = 0.000), left inferior occipital gyrus (t = 3.948, P = 0.000), and right inferior occipital gyrus (t = 3.846, P = 0.000), so it was defined as the occipital lobe hypermetabolism cluster; Cluster 1 showed a decreasing trend in metabolic activity of these regions, thus defined as the non-occipital lobe hypermetabolism cluster. MRI volume analysis revealed that the brain regions with significant volume differences between the non-occipital lobe hypermetabolism cluster and occipital lobe hypermetabolism cluster were the cuneus (t = -2.214, P = 0.038), superior occipital gyrus (t = -2.213, P = 0.038), orbital gyrus (t = -2.829, P = 0.010), lateral part of the superior temporal gyrus (t = -2.246, P = 0.035), occipito-temporal medial sulcus (collateral sulcus), and lingual gyrus sulcus (t = -2.160, P = 0.042). Conclusions Different metabolic patterns of anti-GAD65 antibody-associated encephalitis were identified by ¹⁸F-FDG PET combined with cluster analysis, and their associations with clinical and neuroimaging features were revealed, which deepens the understanding of the pathophysiological mechanisms of this disease.

Key words: Encephalitis, Autoimmune diseases, Glutamate decarboxylase, Positron-emission tomography, Cluster analysis