Ai-Assisted Advanced Mri Using Machine Learning and Deep Learning for Focal Liver Lesion Diagnosis: A Systematic Review Publisher



Validad A ; Koozari A ; Abedi I ; Elhaie M
Source: Indian Journal of Radiology and Imaging Published:2026

Abstract

Background Focal nodular hyperplasia (FNH) and other focal liver lesions (FLLs) present diagnostic challenges due to their diverse imaging characteristics. While the review encompasses FLLs broadly, FNH-specific literature is sparse, with only one study providing dedicated findings, highlighting a key research gap. Advanced magnetic resonance imaging (MRI), including contrast-enhanced and multi-sequence protocols, is critical for accurate diagnosis, but interpretation is labor-intensive and prone to variability. Artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), offers potential to enhance diagnostic accuracy and efficiency in MRI-based FLL and FNH diagnosis. This systematic review evaluates the diagnostic performance, methodologies, and limitations of AI-assisted advanced MRI for FNH and FLLs. Materials and Methods Following PRISMA guidelines, we searched PubMed, Embase, Scopus, and Web of Science up to June 15, 2025, for studies utilizing ML or DL in advanced MRI (e.g., Gd-EOB-DTPA-enhanced, multi-sequence protocols) for FNH and FLL diagnosis. Inclusion criteria encompassed human studies reporting diagnostic metrics (e.g., sensitivity, specificity, AUC). Two reviewers independently screened studies, extracted data, and assessed methodological quality using the QUADAS-2 tool and evaluated reporting completeness using TRIPOD þ AI and CLAIM 2024. Narrative synthesis was performed due to methodological heterogeneity. Results Seven studies met the inclusion criteria, involving diverse AI models (e.g., CNNs, U-Net, DenseNets) and MRI protocols. AI demonstrated high diagnostic performance, with sensitivities of 0.60 to 1.00 and AUCs of 0.83 to 0.99 for FLL detection and classification. FNH-specific findings were limited, with one study reporting a Dice score of 0.777 for FNH segmentation. AI outperformed or matched radiologists, particularly for small lesions, but lacked external validation and standardized reference standards. Common limitations included small datasets and phase-dependent performance. Conclusion AI-assisted advanced MRI shows promise for improving FLL and FNH diagnosis, yet methodological heterogeneity and limited FNH-specific research necessitate cautious interpretation. Future studies should prioritize standardized protocols, larger cohorts, and external validation to enhance clinical applicability.With only seven included studies, this review has scoping elements and limited generalizability, emphasizing the nascent state of the field. © 2026. Indian Radiological Association. All rights reserved.