Mislabeled learning for psychiatric disorder detection

Abstract

The misdiagnosis between schizophrenia (SCZ) and bipolar disorder (BPD) has been a challenge in psychiatry. It is also a long-time unsolved mislabeled learning problem in machine learning and AI fields. In this study, we propose a psychiatric map (pMAP) diagnosis, which is built upon a novel feature self-organizing map (fSOM) algorithm, to tackle it. The psychiatric map summarizes the latent essential characteristics of each observation on a two-dimensional fSOM plane. It solves the misdiagnosis problem by providing high-accuracy psychiatric detection via automatically mislabeled observation identification. Furthermore, pMAP provides powerful and informative visualization for each observation in unveiling hidden psychiatric subtype discovery. This study also presents new insight into the pathology of psychiatric disorders by constructing the devolution path of psychiatric states via relative entropy analysis that discloses latent internal transfer and devolution road maps between different subtypes of the control, BPD, and SCZ groups. To the best of our knowledge, it is the first study to solve mislabel learning for high-dimensional data in machine learning and will inspire more future work in this field.