Diagnosis and surgical decision-making in hepatic hemangioma: From venous malformation overlap to artificial intelligence-assisted multimodality imaging

Table 4 Representative artificial intelligence and radiomics studies relevant to hepatic hemangioma differential diagnosis

Ref.	Task and modality	Main message	Key limitation
Popa et al[20], 2023	Systematic review of automated focal liver lesion diagnosis across US/CT/MRI	Automated diagnosis is feasible, but no modality or model family clearly dominates	Heterogeneous studies; many older single-center datasets
Arribas Anta et al[50], 2025	Narrative review of AI for focal hepatic lesion detection/characterization	AI is strongest as decision support for lesion characterization and workflow assistance	Mostly retrospective evidence; limited real-world implementation data
Pomohaci et al[51], 2025	Systematic review of AI in liver imaging	Liver lesions are the commonest AI target, but external validation and code sharing remain uncommon	Broad liver scope; not restricted to HH-specific questions
Park et al[21], 2026	T2-weighted MRI radiomics for small metastasis vs hemangioma	Excellent discrimination, including visually indeterminate lesions in validation sets	Retrospective study; validation still scanner- and center-dependent
Ozaki et al[22], 2026	Deep-learning reconstruction for DWI differentiation of metastasis vs hemangioma	Improved image quality and lesion conspicuity; diagnostic accuracy similar to standard reconstruction	Small sample size; single-vendor technical ecosystem
Starmans et al[23], 2024/Yang et al[52], 2025	T2-based radiomics for benign vs malignant liver lesions and blood-rich lesion classification	Quantitative T2 features can complement conventional reading when blood-rich benign lesions mimic malignancy	Broader lesion groups rather than pure HH-focused cohorts

HH: Hepatic hemangioma; AI: Artificial intelligence; US: Ultrasound; CT: Computed tomography; MRI: Magnetic resonance imaging; DWI: Diffusion weighted imaging.