Fig. 2: Distribution of MADRS scores across patient transcript data and synthetic data.
From: Using a fine-tuned large language model for symptom-based depression evaluation
The first item (“apparent sadness”) was excluded from the analysis as its rating depends not only on language but also on body language, posture, and facial mimics. Patient transcript data are shown in purple, and synthetic data is shown in light teal. Barplots represent counts of unique scores per item and subject. Abbreviations: MADRS: Montgomery-Åsberg Depression Rating Scale.
