Yu-Hang Liu, Si-Yan Li, Shi-Tian Zhang, The Second Clinical College, Chongqing Medical University, Chongqing 401331, China
Zi-Xuan Zhu, College of Basic Medicine, Chongqing Medical University, Chongqing 400010, China
Author contributions: Liu YH conceived and designed the review, developed the ACTION framework, coordinated the literature search and evidence synthesis, and drafted the manuscript; Li SY contributed to the literature search, evidence screening, data extraction, and manuscript preparation; Zhang ST contributed to evidence verification, preparation of the tables and figures, and critical revision of the manuscript; Zhu ZX contributed to the methodological design, interpretation of the evidence, and critical revision of the manuscript; and all authors reviewed and approved the final version of the manuscript, agreed to be accountable for all aspects of the work.
AI contribution statement: OpenAI Codex (GPT-5) was used as assistive technology for language refinement, structural editing, citation-format checking and manuscript formatting. The authors reviewed and verified all AI-assisted outputs and accept full responsibility for the accuracy, integrity and originality of the manuscript. AI tools were not used to generate data, perform statistical analyses or make autonomous scientific conclusions.
Conflict-of-interest statement: The authors declare no conflict of interest.
Corresponding author: Yu-Hang Liu, The Second Clinical College, Chongqing Medical University, No. 61 University Town Middle Road, Shapingba District, Chongqing 401331, China. 13990562909@163.com
Received: July 22, 2026 Revised: July 29, 2026 Accepted: September 9, 2026 Published online: November 27, 2026 Processing time: 120 Days and 8.2 Hours
Abstract
Metabolic dysfunction-associated steatohepatitis (MASH) has entered a therapeutic era in which priorities extend beyond detecting steatosis to identifying treatment-actionable fibrosis, rationalising referral and biopsy, standardising trial endpoints and monitoring response. Artificial intelligence (AI) spans electronic health records, non-invasive tests, imaging and digital pathology, but reports often omit the user, threshold and consequence. We conducted a structured narrative review and developed the author-proposed ACTION framework. ACTION comprises attached action, calibration in context, threshold-specific benefit, integration, oversight, and non-discrimination and monitoring. Evidence is uneven. Case-finding and enrichment models can identify uncoded or higher-risk patients, but their prospective effects on testing, referral and outcomes remain unproven. Imaging AI supports reproducible quantification, although signals may add little beyond clinical variables. The human-reviewed AI-based measurement tool for MASH (in regulatory records) is among the most mature applications reviewed, within a narrowly specified clinical-trial context. It is qualified for specified MASH clinical-trial workflows, not for autonomous diagnosis or unrestricted treatment selection. AI becomes clinically meaningful only when a calibrated output is linked to a predefined action, an accountable human user and a measurable patient or trial outcome.
Core Tip: Therapeutic advances have shifted the relevant question for artificial intelligence in metabolic dysfunction-associated steatohepatitis from disease detection to decision support. The ACTION framework links each output to an attached action, contextual calibration, threshold-specific benefit, workflow integration, human oversight, and non-discrimination with lifecycle monitoring. Evidence is most mature for human-reviewed digital pathology in specified clinical-trial workflows. Case finding, non-invasive enrichment, biopsy stewardship and response modelling still require prospective evidence that model-guided actions improve care, rather than discrimination alone.