Correlation of histologic, imaging, and artificial intelligence features in NAFLD patients, derived from Gd-EOB-DTPA-enhanced MRI: a proof-of-concept study - Report - MDSpire
Advertisement
Correlation of histologic, imaging, and artificial intelligence features in NAFLD patients, derived from Gd-EOB-DTPA-enhanced MRI: a proof-of-concept study
AI and MRI Features Differentiate NASH from Simple Steatosis in NAFLD Patients
Overview
This study demonstrates that a hybrid AI approach using Gd-EOB-DTPA-enhanced MRI and chemical shift imaging can distinguish non-alcoholic steatohepatitis (NASH) from simple steatosis in NAFLD patients. The AI-derived texture patterns showed predictive value comparable to established MRI metrics and histopathology.
Background
Non-alcoholic fatty liver disease (NAFLD) encompasses a spectrum from simple steatosis to non-alcoholic steatohepatitis (NASH), which can progress to severe liver complications. Liver biopsy remains the gold standard for NASH diagnosis but is invasive and limited by sampling errors. Conventional imaging and serum markers lack specificity to differentiate NASH from simple steatosis. Multiparametric MRI techniques, including Gd-EOB-DTPA-enhanced imaging and chemical shift imaging (CSI), combined with artificial intelligence (AI), offer a promising non-invasive alternative for accurate diagnosis.
Data Highlights
Parameter
Derivation Group (n=46)
Validation Group (n=30)
Age (years)
Not specified
Not specified
Histological Diagnosis
Simple Steatosis or NASH confirmed by biopsy
Simple Steatosis or NASH confirmed by biopsy
MRI Scanner
3-T Magnetom Trio
3-T Magnetom Prisma Fit
Contrast Agent Dose
0.025 mmol/kg Gd-EOB-DTPA
0.025 mmol/kg Gd-EOB-DTPA
AI Technique
Hybrid unsupervised and supervised deep learning (UDC)
Applied model from derivation group
Key Findings
The hybrid AI approach using deep clustering networks (UDC) identified imaging texture patterns predictive of NASH versus simple steatosis.
Relative liver enhancement (RLE) from Gd-EOB-DTPA MRI and fat fraction (FF) from CSI were effective in differentiating NASH from simple steatosis.
AI-derived imaging features correlated with histopathological SAF scores, the gold standard for diagnosis.
The model developed in the derivation cohort was successfully applied to a validation cohort scanned on different MRI equipment and software.
Non-invasive MRI combined with AI may overcome limitations of liver biopsy, including invasiveness and sampling variability.
Clinical Implications
This study supports the use of Gd-EOB-DTPA-enhanced MRI combined with AI-based texture analysis as a non-invasive diagnostic tool to differentiate NASH from simple steatosis in NAFLD patients. Such approaches could reduce reliance on liver biopsy, enabling safer and more acceptable longitudinal monitoring. Integration of these imaging biomarkers into clinical practice may facilitate earlier diagnosis and tailored management of patients at risk for progressive liver disease.
Conclusion
The proof-of-concept investigation demonstrates that AI-enhanced multiparametric MRI can effectively distinguish NASH from simple steatosis, correlating well with histopathology. This non-invasive method holds promise for improved diagnosis and monitoring in NAFLD.
References
European and American Guidelines on NAFLD/NASH Diagnosis and Management
Perkonigg et al. 2021 -- Unsupervised Predictive Texture Discovery in NAFLD Imaging
by Nina Bastati, Matthias Perkonigg, Daniel Sobotka, Sarah Poetter-Lang, Romana Fragner, Andrea Beer, Alina Messner, Martin Watzenboeck, Svitlana Pochepnia, Jakob Kittinger, Alexander Herold, Antonia Kristic, Jacqueline C. Hodge, Stefan Traussnig, Michael Trauner, Ahmed Ba-Ssalamah, Georg Langs
