Multimodal artificial intelligence models for liver fibrosis staging: a scoping review

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Hepatology

Background:

• Multimodal artificial intelligence (AI) integrating heterogeneous data is emerging for liver fibrosis assessment.
• Existing evidence on multimodal AI for liver fibrosis staging is limited by methodological gaps.

Purpose of the Study:

• To comprehensively map evidence on multimodal AI models for liver fibrosis stage prediction.
• To identify common approaches and assess the performance and limitations of these models.

Main Methods:

• Scoping review following Joanna Briggs Institute methodology and PRISMA-ScR guidelines.
• Searched MEDLINE, Web of Science, CENTRAL, and IEEE Xplore for studies using AI/ML models integrating imaging with other data for fibrosis prediction.
• Included 21 studies with 34 distinct multimodal AI models.

Main Results:

• Research is concentrated in China, focusing on hepatitis B-related liver disease.
• CT-based radiomics with serum biomarkers was the most common approach; deep learning was less frequent.
• Median AUC was 0.890, with external validation AUCs ranging 0.808-0.990, but only 20.6% of models reported external validation.

Conclusions:

• The field of multimodal AI for liver fibrosis is nascent with promising diagnostic performance.
• Substantial gaps exist in external validation, calibration reporting, and clinical utility assessment.
• Future research must prioritize rigorous validation and evaluation of clinical decision-making impact.