Plausibility of a Neural Network Classifier-Based Neuroprosthesis for Depression Detection via Laughter Records

Area of Science:

• Computational neuroscience
• Clinical psychology
• Biomedical engineering

Background:

• Current depression diagnostics often rely on subjective assessments or complex physiological signals.
• Analyzing vocalizations, like laughter, offers a non-invasive window into emotional and cognitive states.
• Previous research has not extensively utilized simple acoustic laughter features for depression detection.

Purpose of the Study:

• To evaluate the diagnostic performance of neural network classifiers for depression using laughter sound structures.
• To introduce a novel methodology employing basic acoustic laughter variables as input features.
• To explore the potential of laughter analysis in identifying neurocognitive aspects of depression.

Main Methods:

• Collected 934 laughter samples from 30 patients with depression and 20 healthy controls.
• Extracted elementary sound variables: timing, fundamental frequency mean, formants, average power, and Shannon-Wiener entropy.
• Trained and tested four different neural network models for depression detection using these acoustic features.

Main Results:

• Two neural networks achieved high diagnostic discrimination rates of 93.02% and 91.15%.
• The other two models demonstrated success rates of 87.96% and 82.40%.
• Shannon-Wiener entropy emerged as a critical variable for differentiating between patients and controls.

Conclusions:

• Neural network analysis of laughter acoustics provides a highly effective method for depression diagnosis.
• This approach offers a promising, non-invasive tool for mental health assessment and potentially other neuropsychiatric disorders.
• The findings highlight the neurocognitive link between laughter characteristics and depression, opening avenues for early detection and prognosis research.