Objective:

To develop a multimodal framework for classifying depressive-spectrum states and estimating severity using eye tracking, facial behavior, and language analysis, while effectively handling missing modalities.

Approach:

Key Findings:

• Baseline-3+ achieved accuracy, balanced accuracy, and F1-macro scores approaching 0.90 for classification.
• Lower expected calibration error (specific metric needed) was observed in Baseline-3+ compared to Baseline-3.
• Facial features were the dominant signal for classification, supplemented by eye tracking and language contributions.

Interpretation:

The framework effectively addresses limitations of previous models by supporting classification, severity estimation, and handling of missing modalities, thus enhancing clinical applicability.

Limitations:

• The study primarily focused on a controlled task battery, which may not fully represent real-world scenarios.
• Misclassification was concentrated near the boundary of subthreshold depression.
• The controlled environment may limit the generalizability of findings to broader populations.

Conclusion:

The multimodal framework has potential to augment clinical assessment, particularly for boundary states like subthreshold depression.

Attribution Notice

This content is an AI-generated, fully rewritten summary based on a published scholarly article. It does not reproduce the original text and is not a substitute for the original publication. Readers are encouraged to consult the source for full context, data, and methodology.