Machine Learning and Natural Language Processing in Mental Health: Systematic Review.

Area of Science:

• Artificial Intelligence in Medicine
• Computational Psychiatry
• Digital Health

Background:

• Machine learning (ML) systems and Natural Language Processing (NLP) are AI subfields that learn from data for improved decision-making.
• NLP excels in statistical tasks like text classification and sentiment analysis using large text corpora.
• These technologies are increasingly relevant in medical research and clinical applications.

Purpose of the Study:

• To systematically review and characterize studies employing ML and NLP techniques in mental health research.
• To evaluate the methodological and technical aspects of these studies.
• To consider the potential integration of ML and NLP into mental health clinical practice.

Main Methods:

• Systematic review adhering to PRISMA guidelines and registered with PROSPERO.
• Searches conducted across PubMed, Scopus, ScienceDirect, and PsycINFO using keywords related to ML, data mining, and mental health.
• Inclusion criteria focused on studies in English, excluding case studies, conference papers, and reviews; no date limitations were imposed.

Main Results:

• 58 out of 327 identified articles were included, analyzed qualitatively.
• Studies focused on patients from medical databases, emergency rooms, or social media.
• Key objectives included symptom extraction, severity classification, therapy effectiveness comparison, and psychopathological insights.
• Medical records and social media were primary data sources; Python was the most common platform.

Conclusions:

• ML and NLP are emerging paradigms in medical research, offering new perspectives on unexplored patient data.
• Current applications often confirm existing hypotheses; challenges include imprecise cohorts (e.g., social media users) and language specificity.
• Ethical considerations are paramount before integrating these tools into mental health care; they are best viewed as supportive tools for clinicians.