InterMind: A Doctor-Patient-Family Interactive Depression Assessment System Empowered by Large Language Models

TL;DR

This work tackles the inefficiencies and limited interpretability of depression assessment by introducing InterMind, a doctor-patient-family interactive system powered by large language models. It couples an AI Psychological Chatbot for counseling with an AI Psychologist that generates standardized, DSM-V-aligned assistive diagnostic reports using retrieval-augmented generation and chain-of-thought prompts, guided by clinician oversight. Through data augmentation from social-media dialogues and instruction fine-tuning, the approach improves depression binary classification, severity estimation, report quality, and counseling competency, as validated by clinicians and quantitative metrics. The framework enables continuous monitoring and family involvement, aiming to enhance diagnostic precision, efficiency, and caregiving support in real-world clinical workflows.

Abstract

Depression poses significant challenges to patients and healthcare organizations, necessitating efficient assessment methods. Existing paradigms typically focus on a patient-doctor way that overlooks multi-role interactions, such as family involvement in the evaluation and caregiving process. Moreover, current automatic depression detection (ADD) methods usually model depression detection as a classification or regression task, lacking interpretability for the decision-making process. To address these issues, we developed InterMind, a doctor-patient-family interactive depression assessment system empowered by large language models (LLMs). Our system enables patients and families to contribute descriptions, generates assistive diagnostic reports for doctors, and provides actionable insights, improving diagnostic precision and efficiency. To enhance LLMs' performance in psychological counseling and diagnostic interpretability, we integrate retrieval-augmented generation (RAG) and chain-of-thoughts (CoT) techniques for data augmentation, which mitigates the hallucination issue of LLMs in specific scenarios after instruction fine-tuning. Quantitative experiments and professional assessments by clinicians validate the effectiveness of our system.

Figures (16)

• Figure 1: The framework of proposed LLM-empowered doctor-patient-family interactive depression assessment system. Patients and family members can chat with the AI Psychological Chatbot to share their experiences and feelings. The AI Psychologist then analyzes the dialogue content to generate an assistive depression diagnosis report, offering tailored treatment and caregiving advice to the patient and the family. Additionally, the AI Psychologist can conduct periodic analyses of the mental state based on reports over several days. The doctor reviews the dialogues and reports provided by the system, revising the reports and advice for the patient and family. Patients and families can provide effective feedback of treatment to the system, aiding doctors in optimizing treatment plans.
• Figure 2: Patient and family's user interface, where a,b, and c are patient's chat, report, and treatment strategy pages, while d,e, and f are family's chat, report, and care advice pages, respectively. Here we present the English version of the system's functionality, while other language versions can be found in the Appendix \ref{['appendix1']}.
• Figure 3: Doctor's user interface, which involves patient management menu, timeline, dialogues of patient and family, patient's treatment strategy, family's care advice, and diagnostic report.
• Figure 4: Cyclical analysis interface, which involves the interaction statistics for a month, such as login frequency, chat turns, the distribution of mental states, average scores, and the LLM's analysis of mental states for the month.
• Figure 5: The construction process of AI Psychological Chatbot. The proposed psychological counseling dialogue prompt engineering rewrites the rich experiences and feelings in social media posts {$p_i$} into psychological counseling dialogues {$d_i$}. The generated dialogues $\mathcal{D}_1$ combined with clinical interview dialogues $\mathcal{D}_2$, form a dataset $\mathcal{D}$ used for instruction fine-tuning LLM, ultimately building an AI Psychological Chatbot ($\theta_1^*$).