Scholarlink

<p>Mental health diagnostics have been significantly spurred on by artificial intelligence (AI), particularly for anxiety disorder identification at a young age. However, existing work is fragmented, information is descriptive, and methodological strength is modest. This systematic analysis of AI-based anxiety identification and diagnosis technologies published between 2012 and 2024 synthesizes their functionality, determines their principal methodological and translational gaps, and offers a conceptual taxonomy for future efforts to build upon as well as a mathematical modeling section that formalizes common blending protocols, regularization, and validation pipes. According to PRISMA guidelines, a comprehensive search of PubMed, IEEE Xplore, Google Scholar, and ScienceDirect were performed. After removing duplication and screening on Rayyan, a total of 255 studies were extracted, of which 54 studies were incorporated. Risk of bias and quality appraisal of each study were performed by utilizing the tool PROBAST. A conceptual taxonomy of data modality (EEG, ECG, fMRI, behavioral/text, multimodal), model class (deep learning vs. classical ML), origin of dataset (clinical, laboratory, or crowd-sourced), and clinical application (screening, diagnosis, or monitoring) were built to generate a conceptual taxonomy. Results suggest EEG- and multimodal-based approaches are generally strong on a sample (accuracy ranging from 62.56% to 100%) but are plagued by minuscule cohorts, scarce external verification, and inconsistent reportage, which bode ill for generalizability. Text- and smartphone-based models also suffer from problems of privacy, selectivity, and field deployability. Few of the studies provided mathematical, theoretical, or logical rationale for model selection. Beyond summarization, this review contributes a unified taxonomy, consolidated comparison tables, and a structured quality assessment that clarify methodological strengths, weaknesses, and implementation barriers. It highlights critical research directions including standardized reporting protocols, larger multimodal datasets, ethical AI frameworks, and clinically validated deployment pipelines. By integrating descriptive evidence with conceptual analysis, this review provides a roadmap toward trustworthy, explainable, and clinically applicable AI-driven anxiety diagnostics.</p>

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

AI-based tools for early detection and accurate diagnosis of anxiety disorder (systematic review)

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Asmaa Alaaeldin,
  • Reda Alhajj

摘要

Mental health diagnostics have been significantly spurred on by artificial intelligence (AI), particularly for anxiety disorder identification at a young age. However, existing work is fragmented, information is descriptive, and methodological strength is modest. This systematic analysis of AI-based anxiety identification and diagnosis technologies published between 2012 and 2024 synthesizes their functionality, determines their principal methodological and translational gaps, and offers a conceptual taxonomy for future efforts to build upon as well as a mathematical modeling section that formalizes common blending protocols, regularization, and validation pipes. According to PRISMA guidelines, a comprehensive search of PubMed, IEEE Xplore, Google Scholar, and ScienceDirect were performed. After removing duplication and screening on Rayyan, a total of 255 studies were extracted, of which 54 studies were incorporated. Risk of bias and quality appraisal of each study were performed by utilizing the tool PROBAST. A conceptual taxonomy of data modality (EEG, ECG, fMRI, behavioral/text, multimodal), model class (deep learning vs. classical ML), origin of dataset (clinical, laboratory, or crowd-sourced), and clinical application (screening, diagnosis, or monitoring) were built to generate a conceptual taxonomy. Results suggest EEG- and multimodal-based approaches are generally strong on a sample (accuracy ranging from 62.56% to 100%) but are plagued by minuscule cohorts, scarce external verification, and inconsistent reportage, which bode ill for generalizability. Text- and smartphone-based models also suffer from problems of privacy, selectivity, and field deployability. Few of the studies provided mathematical, theoretical, or logical rationale for model selection. Beyond summarization, this review contributes a unified taxonomy, consolidated comparison tables, and a structured quality assessment that clarify methodological strengths, weaknesses, and implementation barriers. It highlights critical research directions including standardized reporting protocols, larger multimodal datasets, ethical AI frameworks, and clinically validated deployment pipelines. By integrating descriptive evidence with conceptual analysis, this review provides a roadmap toward trustworthy, explainable, and clinically applicable AI-driven anxiety diagnostics.