Detection of major depressive disorder in adolescents based on textual and acoustic features

Adolescent major depressive disorder (MDD) has become a growing public health concern, with rising prevalence over the past decade and significant impacts on social, academic, and occupational functioning. However, diagnosing adolescent MDD is hindered by limited specialist access and time-consuming gold-standard interviews. Therefore, it is imperative to develop an adolescent MDD detection system to provide auxiliary diagnosis and facilitate accurate clinical assessment by physicians.

We developed the Intelligent Mind Chatroom (IMC), which adopts four structured tasks (positive, negative, neutral interviews; text reading), to facilitate standardized speech data collection. A total of 131 adolescents aged 13-18 years were enrolled: 112 participants (56 MDD patients and 56 healthy controls) in the internal validation cohort, and 19 MDD patients in the external validation cohort. We extracted and analyzed textual and acoustic features, proposed a hybrid fusion method for classification, and conducted external validation.

The proposed hybrid fusion model demonstrated effective detection performance, achieving an accuracy of 0.91, recall of 0.91, F1 score of 0.91, AUROC of 0.96, AUPRC of 0.97, and a Brier score of 0.09 on internal evaluation, and a sensitivity of 0.90 on an external validation set.

The primary limitation of this study is the relatively small sample size, which may constrain the generalizability of the model.

Detecting MDD in adolescents based on textual and acoustic features demonstrates potential as an objective tool. The proposed hybrid fusion method provides an experimental foundation for future research and facilitates practical implementation in clinical settings.