Key Findings:
- Major depressive disorder affects over 280 million people worldwide, making early detection crucial for effective treatment.
- The study utilized machine learning models to classify individuals based on their WhatsApp voice messages.
- Two datasets were employed: one for training the LLMs and another for testing them.
Study Details:
- The training dataset included 86 participants: 37 women and 8 men diagnosed with major depressive disorder, alongside a control group of 41 volunteers.
- The testing dataset comprised 74 participants: 33 outpatients diagnosed with major depressive disorder and 41 control group participants.
- All participants provided informed consent and were screened for confounding factors.
Performance Insights:
- The LLM demonstrated higher accuracy in classifying women compared to men, particularly with the “describe your week” audio, achieving 91.9% accuracy for women.
- For male participants, the accuracy was 75% for the same audio type.
- When analyzing “count to 10” data, the model’s accuracy was 82% for women and 78% for men.
Future Implications:
The authors are optimistic that further refinement of these models could lead to a cost-effective and practical method for screening individuals for depression, as well as other clinical and research applications. Senior author Lucas Marques stated, “Our study shows that subtle acoustic patterns in spontaneous WhatsApp voice messages can help identify depressive profiles with surprising accuracy using machine learning.”
This research opens a promising avenue for developing low-burden, real-world digital screening tools that align with people’s daily communication habits.
For more details, access the full article in PLOS Mental Health: ML-based detection of depressive profile through voice analysis in WhatsApp™ audio messages of Brazilian Portuguese Speakers.