Visualizing diagnosis-associated AMPA receptor-related [11C]K-2 PET SUVR patterns across psychiatric disorders using weakly supervised UMAP

Abstract

AMPA receptors (AMPARs) mediate fast excitatory neurotransmission and play a central role in synaptic function, with accumulating evidence in psychiatric disorders. Using [ 11 C]K-2 PET to assess AMPAR-related signals in vivo, we examined whether regional standardized uptake value ratio (SUVR) patterns show diagnosis-aligned neighborhood organization beyond label-permutation chance. We analyzed the 219-participant cohort previously reported by Hatano et al., comprising healthy controls and individuals with autism spectrum disorder, bipolar disorder, major depressive disorder, or schizophrenia. Data were parcellated into 75-dimensional SUVR vectors using white-matter (WM) referencing (SUVR-WM) or whole-brain normalization (SUVR-WB). Uniform Manifold Approximation and Projection (UMAP) was applied unsupervised and with minimal supervision. Weakly supervised embeddings were compared with 1000 label-permuted embeddings using Calinski–Harabasz Index, Davies–Bouldin Index, and Silhouette scores with Bonferroni correction. Covariate associations were visualized. With minimal supervision, SUVR-WB produced diagnosis-aligned organization among the five groups and large improvements over permuted-label embeddings. SUVR-WM formed a horseshoe manifold aligned with mean SUVR-WM and age, with limited diagnostic clustering; age/sex residualization did not alter this geometry. These exploratory findings suggest that multiregional [ 11 C]K-2 PET SUVR patterns show diagnosis-associated organization within a transdiagnostic sample, while emphasizing that normalization critically shapes visualization and that this approach is not an individual diagnostic test.