Scholarlink

The advancement of Generative Artificial Intelligence (AI), such as Large Language Models (LLMs), has caused a paradigm change in several industries, including healthcare. This work investigates the potential of LLMs, specifically, Meta’s LLaMA3-70B-8192 model. The model is tasked with creating and assessing a potential dataset of question-and-answer (QA) pairs concerning the follow- ing six mental health disorders: Anxiety, Bipolar Disorder, Borderline Personality Disorder, Depression, Panic Disorder, and Schizophrenia. Two methodologies are employed on the preprocessed data. The first methodology assesses the relevance of question and answer pairs by using the transformer-based BERT model to determine their semantic similarity. The second methodology uses XAI techniques like LIME (Local Interpretable Model Agnostic Explanations), which assist in identifying significant fea- tures in text or graphical images and offer interpretations. It also portrays the capability of LLMs to improve diagnostic accuracy, enhance understanding of treatment, and provide explanations for predic- tions thus elevating the confidence level in AI systems. The model’s efficiency was assessed using the Explainable AI (XAI) along with the BERT model, where we used BiomedBERT which recorded the best Precision 0.9398, a Recall 0.9001, and F1 Score 0.9195 compared to ClinicalBERT, and SciBERT. Logistic Regression (LR) outperforms the other classifiers Naive Bayes (NB), and K-nearest neighbors (KNN) with an accuracy of 0.9264 which can further be used with LIME for prediction. These results improve mental health diagnoses and enhance healthcare efficiency using AI.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Leveraging Large Language Models for Mental Health Diagnostics

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Priyanka V. Mahadik,
  • Vinay Vijay Tawde,
  • Jyotshna Dongardive

摘要

The advancement of Generative Artificial Intelligence (AI), such as Large Language Models (LLMs), has caused a paradigm change in several industries, including healthcare. This work investigates the potential of LLMs, specifically, Meta’s LLaMA3-70B-8192 model. The model is tasked with creating and assessing a potential dataset of question-and-answer (QA) pairs concerning the follow- ing six mental health disorders: Anxiety, Bipolar Disorder, Borderline Personality Disorder, Depression, Panic Disorder, and Schizophrenia. Two methodologies are employed on the preprocessed data. The first methodology assesses the relevance of question and answer pairs by using the transformer-based BERT model to determine their semantic similarity. The second methodology uses XAI techniques like LIME (Local Interpretable Model Agnostic Explanations), which assist in identifying significant fea- tures in text or graphical images and offer interpretations. It also portrays the capability of LLMs to improve diagnostic accuracy, enhance understanding of treatment, and provide explanations for predic- tions thus elevating the confidence level in AI systems. The model’s efficiency was assessed using the Explainable AI (XAI) along with the BERT model, where we used BiomedBERT which recorded the best Precision 0.9398, a Recall 0.9001, and F1 Score 0.9195 compared to ClinicalBERT, and SciBERT. Logistic Regression (LR) outperforms the other classifiers Naive Bayes (NB), and K-nearest neighbors (KNN) with an accuracy of 0.9264 which can further be used with LIME for prediction. These results improve mental health diagnoses and enhance healthcare efficiency using AI.