Scholarlink

<p>Trustworthiness remains a pivotal challenge in deploying artificial intelligence (AI) for healthcare decision making, where opaque models risk bias, errors, and ethical lapses. The AI Trustworthiness Index for Healthcare (AITI-H) is a structured, multi-domain assessment tool designed to translate high-level principles for trustworthy artificial intelligence into measurable, operational criteria for healthcare organizations. AITI-H is grounded in a foundational conceptual distinction between trust—a relational and psychological state held by a clinician, patient, or regulator—and trustworthiness, which denotes the objective, verifiable properties of the AI system itself that warrant such trust. By targeting trustworthiness as a measurable construct, AITI-H provides an evidence base for calibrated trust rather than subjective perception alone. Anchored in a systematic synthesis of international guidelines and regulatory instruments, including the NIST AI Risk Management Framework (AI RMF 1.0), the EU AI Act and its 2025 implementation roadmap, and healthcare-specific frameworks such as FUTURE‑AI, AITI-H organizes trustworthiness into five core domains—explainability, fairness, robustness, privacy, and accountability—operationalized through 25 sub‑indicators, each graded across five quantitative or semi‑quantitative thresholds. To prevent high scores in one domain from masking critical deficiencies elsewhere, non-compensatory scoring rules are applied. Implementation feasibility is supported through a three-tier resource stratification (Basic, Intermediate, Advanced) that allows organisations to apply the framework proportionately to their governance maturity and technical capacity. The overarching aim of AITI-H is to provide healthcare organizations, regulators, and developers with a practical, reproducible index that captures both technical and governance dimensions of AI trustworthiness, especially for high‑risk clinical applications regulated as medical devices or high‑risk AI systems. By adhering to a standard tool development protocol grounded in consensus methods using modified Delphi method and integrating a multidisciplinary consultation experts, AITI-H is designed not only as a checklist, but as a robust measurement instrument capable of benchmarking systems over time and across settings.</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 Trustworthiness Index for Healthcare (AITI-H): conceptualization, structure, and development

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Hanif Abdul Rahman,
  • Amirul Ariffin Noraidi,
  • Hein Minn Tun

摘要

Trustworthiness remains a pivotal challenge in deploying artificial intelligence (AI) for healthcare decision making, where opaque models risk bias, errors, and ethical lapses. The AI Trustworthiness Index for Healthcare (AITI-H) is a structured, multi-domain assessment tool designed to translate high-level principles for trustworthy artificial intelligence into measurable, operational criteria for healthcare organizations. AITI-H is grounded in a foundational conceptual distinction between trust—a relational and psychological state held by a clinician, patient, or regulator—and trustworthiness, which denotes the objective, verifiable properties of the AI system itself that warrant such trust. By targeting trustworthiness as a measurable construct, AITI-H provides an evidence base for calibrated trust rather than subjective perception alone. Anchored in a systematic synthesis of international guidelines and regulatory instruments, including the NIST AI Risk Management Framework (AI RMF 1.0), the EU AI Act and its 2025 implementation roadmap, and healthcare-specific frameworks such as FUTURE‑AI, AITI-H organizes trustworthiness into five core domains—explainability, fairness, robustness, privacy, and accountability—operationalized through 25 sub‑indicators, each graded across five quantitative or semi‑quantitative thresholds. To prevent high scores in one domain from masking critical deficiencies elsewhere, non-compensatory scoring rules are applied. Implementation feasibility is supported through a three-tier resource stratification (Basic, Intermediate, Advanced) that allows organisations to apply the framework proportionately to their governance maturity and technical capacity. The overarching aim of AITI-H is to provide healthcare organizations, regulators, and developers with a practical, reproducible index that captures both technical and governance dimensions of AI trustworthiness, especially for high‑risk clinical applications regulated as medical devices or high‑risk AI systems. By adhering to a standard tool development protocol grounded in consensus methods using modified Delphi method and integrating a multidisciplinary consultation experts, AITI-H is designed not only as a checklist, but as a robust measurement instrument capable of benchmarking systems over time and across settings.