Single and Cross-Disorder Detection for Autism and Schizophrenia

Abstract Detection of mental disorders from textual input is an emerging field for applied machine and deep learning methods. Here, we explore the limits automated detection autism spectrum disorder (ASD) schizophrenia (SCZ). We compared performance of: (1) dedicated diagnostic tools that involve collecting data, (2) methods to data gathered by these tools, (3) psychiatrists. Our article tests effectiveness several baseline approaches, such as bag words dictionary-based vectors, followed a model. employed two more refined Sentic text representations using affective features concept-level analysis on texts. Further, selected state-of-the-art representation inference, well experimented with transfer zero-shot learning. Finally, also explored few-shot low size scenarios, which typical problem clinical setting. The best breed outperformed human raters (psychiatrists). Cross-dataset approaches turned out be useful (only SCZ ASD) despite different types. revealed promising results dataset. However, effort needed efficiently training models, given very limited amounts labeled data. Psychiatry one few medical fields in diagnosis most based subjective assessment psychiatrist. Therefore, introduction objective supporting diagnostics seems pivotal. This paper step this direction.