A Nearest Neighbor Propagation-Based Partial Label Learning Method for Identifying Biotypes of Psychiatric Disorders

Diagnoses of psychiatric disorders only based on clinical presentation are less reliable. In clinical practice, it is difficult to distinguish bipolar disorder with psychosis (BPP), schizoaffective disorder (SAD), and schizophrenia (SZ) as they have many overlapping symptoms. Therefore, there is an urgent need to develop new methods to help increase diagnostic reliability or even explore biotypes for the psychiatric disorders by using neuroimaging measures such as brain functional connectivity (FC). Partial label learning can extract valid information from subjects with incompletely accurate labels, however it has not been well studied in the neuroscience field. Here, we propose a new partial label learning method to explore transdiagnostic biotypes using FC estimated from functional magnetic resonance imaging (fMRI) data. Our method iteratively mines reliable information from available subjects and then propagates the gained knowledge in a typical $$K+N$$ graph structure. Based on fMRI data from 113 BPP patients, 113 SAD patients, 113 SZ patients, and 113 healthy controls (HC), meaningful biotypes are obtained using our method, showing significant differences in FC. In conclusion, the proposed method is promising in extracting biotypes of psychiatric disorders.