An insight into diagnosis of depression using machine learning techniques: a systematic review

Area of Science:

• Mental Health Research
• Computational Psychiatry
• Biomedical Informatics

Background:

• Depression is a prevalent mental disorder with heterogeneous symptoms, often co-occurring with other conditions.
• Current depression diagnosis relies heavily on clinician expertise.
• Machine learning and diverse data modalities are increasingly used to aid depression identification.

Purpose of the Study:

• To analyze trends in publications, data modalities, and machine learning models for depression diagnosis.
• To identify prevalent pre-processing and feature selection techniques.
• To guide future research directions in automatic depression diagnosis.

Main Methods:

• Systematic review of articles from IEEE Xplore and PubMed (2011-April 2021).
• Analysis of 135 selected articles based on defined inclusion criteria.
• Statistical analysis using one-way ANOVA and Tukey-Kramer test.

Main Results:

• Significant growth in publications related to depression diagnosis research.
• High diversity in data modalities and machine learning classifiers used.
• Functional MRI data with Support Vector Machine classifiers identified as a popular combination.
• Data scarcity and small sample sizes are key challenges, especially for neuroimaging data.

Conclusions:

• Machine learning techniques combined with appropriate data modalities show promise for automatic depression diagnosis.
• Addressing data limitations is crucial for advancing the field.
• Further research can refine diagnostic accuracy and clinical utility.