Speech feature identification model for depressed individuals with suicidal ideation based on autobiographical memory

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Suicidal ideation in depression is a critical predictor of suicide risk, yet early identification is challenging.
• Current machine learning models struggle to differentiate suicidality markers from general severe depression symptoms.
• Objective and early identification of suicide risk in depressed individuals remains a significant clinical challenge.

Purpose of the Study:

• To develop and validate a multimodal model integrating vocal features and autobiographical memory.
• To specifically distinguish depressed patients with suicidal ideation from those without.
• To provide objective markers for early suicide risk identification in depression.

Main Methods:

• 88 depressed patients were grouped by severity and suicidal ideation (mild depression without suicidal ideation, moderate depression with suicidal ideation, severe depression with suicidal ideation).
• Utilized the Autobiographical Memory Test (AMT), clinical scales (BDI-II, OGMQ), and vocal feature extraction.
• Employed Random Forest machine learning models and analysis of variance (ANOVA) for classification and group comparisons.

Main Results:

• Patients with suicidal ideation showed overgeneralized autobiographical memory and distinct vocal patterns (reduced prosody, altered spectral energy).
• A Random Forest model achieved high classification accuracy (AUC up to 1.00).
• Model interpretability (SHAP) indicated autobiographical memory is key for initial suicidal ideation detection, while depression severity is more important for differentiating moderate vs. severe suicidal cases.

Conclusions:

• An integrated analysis of vocal features and autobiographical memory, via an interpretable machine learning model, offers an objective approach for predicting suicidal ideation in depression.
• This multimodal method effectively differentiates patients with and without suicidal ideation, offering novel insights into cognitive and physiological suicide risk markers.
• This represents a significant advancement towards precise, clinically applicable tools for early suicide risk identification and intervention.