Artificial intelligence in metabolic dysfunction-associated steatotic liver disease: Machine learning for non-invasive diagnosis and risk stratification

Table 1 Artificial intelligence -based models for metabolic dysfunction-associated steatotic liver disease diagnosis and their performance

Ref.	AI model	Data used	Key finding	AUC/performance
Yu et al[3], 2025	RF	50 clinical and biochemical features (e.g., BMI, liver enzymes, metabolic markers); validated against histology or elastography in subsets of cohorts	Developed an explainable 10-feature RF model for MASLD prediction	AUC: 0.928 (internal), 0.918 (external)
Wakabayashi et al[8], 2025	SVM	Clinical and laboratory markers (excluding platelet counts); validated against histology or elastography in subsets of cohorts	Predicted fibrosis stages without platelet count or elastography	AUC: 0.886 (≥ F2), 0.916 (cirrhosis)
Byra et al[9], 2022	CNN (ultrasound)	Applied CNNs to ultrasound images for steatosis detection	Deep learning-based fat quantification using transfer learning	AUC: 0.91 (PDFF ≥ 5%)
Yasaka et al[10], 2018	DL (CT)	Utilized CT-based DL for fibrosis staging; referenced biopsy-confirmed fibrosis stages for model validation	Staged liver fibrosis from CT scans (moderate correlation with histopathology)	Moderate performance (needs improvement)

AI: Artificial intelligence; MASLD: Metabolic dysfunction-associated steatotic liver disease; RF: Random forest; BMI: Body mass index; SVM: Support vector machine; CNN: Convolutional neural network; CT: Computed tomography; DL: Deep learning; PDFF: Proton density fat fraction; AUC: Area under curve.