Interventional AI-Human Collaboration for Steatotic Liver Disease Screening

Official title

AI-Driven Opportunistic Screening and Risk-Adapted Management of Steatotic Liver Disease

Overview

Steatotic liver disease (SLD) is one of the most prevalent chronic liver diseases worldwide, affecting nearly 30% of the global population and projected to exceed 55% by 2040. Timely identification and management of intermediate- and high-risk SLD patients are essential, yet early detection remains challenging because current diagnostic modalities, such as biopsy, ultrasonography, and serum indices, are invasive, insensitive, operator-dependent, or difficult to scale. In contrast, non-contrast CT is widely available in routine care and offers substantial potential for opportunistic SLD screening, although this value has not been fully utilized. Our previously developed MAOSS model accurately identifies intermediate- and high-risk individuals, with MAOSS score≥1.6 combined with Fibro Score ≥1.7, demonstrating high sensitivity and specificity in our large-scale retrospective study. However, despite these promising retrospective findings, the model has not undergone prospective interventional validation, and it remains unclear whether an AI-guided workflow can truly enhance clinical risk stratification, diagnostic yield, and downstream management in real-world SLD populations. Therefore, a prospective intervention study is needed to determine whether MAOSS-guided identification and recall of at-risk individuals can meaningfully improve fibrosis detection and optimize clinical care pathways for SLD.

Detailed description

The AIG-SLD Screening Project is a single-arm, open-label, prospective interventional study designed to evaluate the effectiveness of a MAOSS-guided identification and AI-human collaboration recall strategy for detecting individuals at intermediate or high risk of steatotic liver disease (SLD) and for assessing intervention outcomes. The trail will prospectively and consecutively enroll around 8000 eligible adults aged ≥18 years who undergo routine chest or abdominal non-contrast CT (NCCT) with adequate hepatic coverage .

The AI system i.e. MAOSS will be embedded within the standard clinical workflow to evaluate the real-world performance and the impact of AIG-SLD screening. All eligible NCCT scans will be evaluated through two parallel streams:

1. Standard of Care (SoC) workflow: Radiologists perform independent assessments, first-line radiologists review followed by a senior radiologist finalizing the report. 2. AIG workflow: The MAOSS system simultaneously analyzes the identical imaging data in real-time.

The system screens patients with clinically suspected SLD by flagging those with a MAOSS score ≥1.6 and a FIBRO Score ≥1.7 for recall. These algorithmic flags will be compared against radiologists' determinations of clinically significant SLD. Management pathways are defined as follows: (1) Concordant cases: If the Standard of Care (SoC) and the AIG pathway agree (both recommending recall or both recommending no recall), the agreed-upon decision will be executed. (2) Discordant cases: If the SoC and AIG pathways disagree, patients will be recalled for primary hepatology care to ensure safety.

Primary hepatology care begins with the collection of questionnaires regarding medical history, lifestyle, alcohol consumption, and metabolic risks (Type 2 Diabetes Mellitus, obesity, metabolic syndrome, significant alcohol consumption, or viral hepatitis) followed by further serum laboratory tests and Transient Elastography (e.g., FibroScan). Recalled patients will be managed according to the MAOSS pathway starting with FIB-4 stratification (\<1.3, 1.3-2.67, \>2.67), followed by FAST stratification (\<0.35, ≥0.35) as needed. Intermediate-to-high-risk patients will proceed to escalated care involving MRE, MRI-PDFF, or liver biopsy. Management is then determined by MRE-derived Liver Stiffness Measurement (LSM): patients with LSM \< 3.5 kPa will receive lifestyle interventions and annual reassessment; those with LSM 3.5-5.0 kPa (F2-F3) will receive pharmacological or therapeutic interventions; and those with LSM ≥ 5.0 kPa (F4) will undergo cirrhosis-based management. Patients in the latter two groups will be reassessed every six months to monitor changes in steatosis and fibrosis.

Interventions

• Diagnostic test AI-human collaboration for SLD screening
  The system screens patients with clinically suspected SLD by flagging those with a MAOSS score ≥1.6 and a FIBRO Score ≥1.7 for recall. These algorithmic flags will be compared against radiologists' determinations of clinically significant SLD. Management pathways are defined as follows: (1) Concordant cases: If the Standard of Care (SoC) and the AIG pathway agree (both recommending recall or both recommending no recall), the agreed-upon decision will be executed. (2) Discordant cases: If the SoC

Primary outcome measures

• Effective referral yield for escalated hepatology care [Time frame: 6 months]

Secondary outcome measures (8)

• Real-World Screening Specificity [Time frame: Duration of the study (12 months)]
• Real-World Screening Sensitivity [Time frame: Duration of the study (12 months)]
• Real-World Screening Negative predictive value [Time frame: Duration of the study (12 months)]
• Intervention Success (Fibrosis Reversion) [Time frame: Duration of the study (12 months)]
• Intervention Success (Steatosis Reversion) [Time frame: Duration of the study (12 months)]
• Quantitative Liver Stiffness Trends [Time frame: Duration of the study (12 months)]
• Quantitative Liver Steatosis Trends [Time frame: Duration of the study (12 months)]
• Intervention Patient Adherence [Time frame: Duration of the study (12 months)]

Eligibility criteria

Inclusion criteria

• Adults aged ≥18 years undergoing routine non-contrast or contrast-enhanced chest or abdominal CT examination.
• CT images with adequate hepatic coverage and sufficient image quality for MAOSS analysis.
• Willing to undergo the recall evaluation and either: having a FIB-4 result within the past 1 month, or willing to complete blood testing (ALT, AST, platelet count) required for FIB-4 calculation and undergo FibroScan or MRE assessment.
• Willing to participate in the study and able to provide written informed consent at the time of recall.

Exclusion criteria

• Known malignant liver tumors (e.g., HCC, cholangiocarcinoma) or a history of liver transplantation or major hepatic resection.
• Known cirrhosis based on noninvasive fibrosis assessment tests, liver biopsy or complications of decompensated disease, or with a documented history of cirrhosis identified by clinical notes will be excluded.
• Biliary obstruction, acute cholangitis, or other conditions that may interfere with interpretation of liver biochemistry or fibrosis risk assessment.
• CT images with severe artifacts or incomplete liver coverage preventing reliable MAOSS analysis.
• Severe acute systemic illness (e.g., sepsis, shock, acute heart failure), or pregnancy or breastfeeding.
• Unwilling or unable to complete recall procedures, including required blood tests, FibroScan, or MRE when indicated, or unable to comply with study follow-up.
• Severe comorbidity with an expected survival of less than 1 year (e.g., terminal malignancy).

Criteria are shown verbatim from the registry (in English). Final eligibility is always assessed by the study center.

Healthy volunteers: Yes

Study design

Allocation

N/A

Model

Single group

Masking

Open label

Primary purpose

Screening

Study locations

China · 1 center

• Shengjing Hospital of China Medical University — Shenyang

Publications

• Gao Y, Li C, Chang W, Du B, Ye X, Yeo YH, Xia Y, Guo H, Zhang X, Liu W, Bai R, Li B, Hong Y, Yao J, Lu L, Cao K, Yan K, Chen J, Li J, Hou Y, Zhang L, Shi Y. Multi-modal AI for opportunistic screening, staging and progression risk stratification of steatotic liver disease. Nat Commun. 2026 Feb 11;17(1):1562. doi: 10.1038/s41467-026-68414-3. PMID 41672973

Identifiers

NCT: NCT07613827 · SH-CMU-SLD-Intervention

Primary sources (government registries)

• ClinicalTrials.gov (US NIH)

View this study on ClinicalTrials.gov ↗