HepaFat-AI

HepaFat-AI is a software medical device that employs artificial intelligence, specifically convolutional neural networks, to automatically analyze magnetic resonance imaging scans for the quantitative assessment of liver fat. Its purpose is to provide objective and reproducible measurements of volumetric liver fat fraction, proton density fat fraction, and steatosis grade, thereby assisting clinicians in the diagnosis, monitoring, and management of fatty liver disease and in the evaluation of potential liver donors.

HepaFat-AI is an advanced software system designed for the automated quantitative analysis of liver fat from magnetic resonance imaging (MRI) data. It processes DICOM images, acquired via a specific "HepaFat-Scan" protocol, to determine volumetric liver fat fraction (VLFF), proton density fat fraction (PDFF), and steatosis grade. The core technology relies on artificial intelligence, specifically two convolutional neural networks (CNNs): one for predicting an intermediate parameter "Alpha" related to fat content and another for anomaly detection, which minimizes user input and potential human error. Algorithmic modules then convert this AI-derived Alpha parameter into clinically relevant metrics (VLFF, PDFF, steatosis grade) using established lookup tables, based on the MRI principle of differentiating hydrogen nuclei signals in water versus triglyceride fat. The system is indicated for assessing liver fat in individuals with suspected or confirmed fatty liver disease, monitoring liver fat content in patients undergoing weight loss management, and aiding in the assessment and screening of living donors for liver transplant. Performance data, validated against the predicate device (HepaFat-Scan) and liver biopsy results from a cohort of 145 adult and pediatric NAFLD/NASH patients, demonstrated that HepaFat-AI possesses a diagnostic ability comparable to the previous, more manual method. For instance, in distinguishing no steatosis from any steatosis (Grade 0 vs Grades 1-3), HepaFat-AI achieved a sensitivity of 97.6% and a specificity of 88.2%, metrics that were similar to or slightly better than the predicate device when both were compared against biopsy-confirmed steatosis grades.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the 'locked-down' AI model account for potential variations in MRI scanner hardware or acquisition parameters across different clinical sites that might still fall within the 'HepaFat-Scan protocol' but introduce subtle image differences?; What specific image features or characteristics does the anomaly detection convolutional neural network identify, and how are these anomalies flagged or addressed in the clinical workflow?; Could the AI-derived 'Alpha' parameter offer novel diagnostic insights beyond its direct conversion to VLFF, PDFF, and steatosis grade, particularly in early-stage or atypical presentations of liver disease?; Given the performance comparison against liver biopsy, what is the inter-observer and intra-observer variability of HepaFat-AI compared to traditional histological assessment of steatosis, especially for borderline or equivocal cases?; How does the HepaFat-AI system perform in specific patient subpopulations, such as pediatric patients or individuals with co-existing liver conditions like iron overload or fibrosis, and are there any validated adjustments or known limitations for these groups?