Year: 2026
Type: conference-paper
Abstract: Accurate and interpretable predictions of depression severity are essential for clinical decision support, yet existing models often lack uncertainty estimates and temporal interpretability.We propose PTTSD, a Probabilistic framework for Depression Detection from clinical interview utterance sequences that predicts PHQ-8 scores while modeling calibrated uncertainty.PTTSD includes sequence-to-sequence and sequence-to-one variants, both combining LSTMs, self-attention, and residual connections with Gaussian or Student's-t output heads trained via negative log-likelihood.The sequence-to-sequence variant enables temporal analysis of how predictive confidence evolves over an interview, despite the target being a single session-level score.Evaluated on E-DAIC and DAIC-WOZ, PTTSD achieves competitive performance among text-only systems (e.g., MAE = 3.85 on E-DAIC, 3.55 on DAIC) and produces well-calibrated prediction intervals.Ablations confirm the value of attention and probabilistic modeling, while a three-part calibration analysis and qualitative case studies highlight the clinical relevance of uncertaintyaware prediction.E-DAIC E-DAIC seq-to-one (MentalBERT) vs. seq-to-seq (MentalBERT) MAE 4.18 4.20 -0.594 0.591 ns E-DAIC seq-to-one (MentalBERT) vs. seq-to-one (MiniLM) MAE 4.18 4.58 -1.379 0.300 ns E-DAIC seq-to-one (MentalBERT) vs. seq-to-seq (MiniLM) MAE 4.18 3.85 +8.927 0.001 * * E-DAIC seq-to-seq (MentalBERT) vs. seq-to-one (MiniLM) MAE 4.20 4.58 -1.314 0.319 ns E-DAIC seq-to-seq (MentalBERT) vs. seq-to-seq (MiniLM) MAE 4.20 3.85 +12.124 <0.001 * * * E-DAIC seq-to-one (MiniLM) vs. seq-to-seq (MiniLM) MAE 4.58 3.85 +2.521 0.126 ns E-DAIC seq-to-one (MentalBERT) vs. seq-to-seq (MentalBERT) RMSE 5.23 5.39 -1.816 0.158 ns E-DAIC seq-to-one (MentalBERT) vs. seq-to-one (MiniLM) RMSE 5.23 5.87 -1.193 0.351 ns E-DAIC seq-to-one (MentalBERT) vs. seq-to-seq (MiniLM) RMSE 5.23 4.52 +3.062 0.071 † E-DAIC seq-to-seq (MentalBERT) vs. seq-to-one (MiniLM) RMSE 5.39 5.87 -0.900 0.462 ns E-DAIC seq-to-seq (MentalBERT) vs. seq-to-seq (MiniLM) RMSE 5.39 4.52 +3.880 0.053 † E-DAIC seq-to-one (MiniLM) vs. seq-to-seq (MiniLM) RMSE 5.87 4.52 +2.349 0.111 ns
Authors: Fabian Schmidt, Seyedehmoniba Ravan, Vladimir Vlassov
Institutions: KTH Royal Institute of Technology, Uppsala University
Language: English
Topic: Mental Health via Writing
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W4416026361