Discovering the neural correlate informed nosological relation among multiple neuropsychiatric disorders through dual utilisation of diagnostic information

The unravelled nosological relation among diverse types of neuropsychiatric disorders serves as an important precursor in advocating the dimensional approach to psychiatric classification. Leveraging high-dimensional abnormal resting-state functional connectivity, the crux of mining corresponded nosological relations is to derive a low-dimensional embedding space that preserves the diagnostic attributes of represented disorders. To accomplish this goal, we seek to exploit the available diagnostic information in learning the optimal embedding space by proposing a novel type of conditional variational auto-encoder that incorporates dual utilisation of diagnostic information. Encouraged by the achieved promising results in challenging the conventional approaches in low dimensional density estimation of synthetic functional connectivity features, we further implement our approach on two empirical neuropsychiatric neuroimaging datasets and discover a reliable nosological relation among autism spectrum disorder, major depressive disorder, and schizophrenia.