(231 days)
HepaFat-AI is indicated to:
Assess the volumetric liver fat fraction, proton density fat fraction and steatosis grade in individuals with confirmed or suspected fatty liver disease;
When interpreted by a trained physician, the results can be used to
monitor liver fat content in patients undergoing weight loss management and can be used to
aid in the assessment and screening of living donors for liver transplant.
The HepaFat-AI Analysis System is a software platform designed to automatically analyse magnetic resonance imaging (MRI) datasets to generate an estimate of the patient's volumetric liver fat fraction (VLFF). To carry out an analysis, the user simply uploads raw DICOM images to the HepaFat-AI Analysis System. No user input is required for the analysis thus minimising the impact of human error on obtained results. The HepaFat-AI system requires image input data that have been acquired according to the HepaFat-Scan protocol.
The key components for the HepaFat-AI Analysis System for volumetric liver fat fraction measurement are:
Magnetic Resonance Imaging Protocol: The use of a specific magnetic resonance imaging protocol for acquisition of the raw image data. The imaging protocol is critical to ensure the quality of the end results. Its adherence is verified by the IOC Module, an automated algorithm that checks the correctness of each parameter in the protocol.
HepaFat-AI Analysis Software: Custom-designed image analysis software performing the Alpha measurement and anomaly detection based on Artificial Intelligence (AI) technology. It is composed of 2 convolutional neural networks. The primary network is for the prediction of Alpha and a secondary network is for anomaly detection. This element is considered the medical device for a regulatory perspective. Following the training of the AI tool, the system is 'locked-down' for final validation prior to release in commercial use to ensure reproducibility of the results.
Volumetric Liver Fat Fraction Measurement: An additional software module (algorithmic) that incorporates a conversion lookup table relating Alpha to volumetric liver fat fraction (VLFF) is added to allow production of a volumetric liver fat fraction report.
Proton Density Fat Fraction Measurement: An additional software module (algorithmic) that incorporates a conversion lookup table relating VLFF to proton density fat fraction (PDFF) is added to allow production of a proton density fat fraction report.
Steatosis Grade Measurement: An additional software module (algorithmic) that incorporates a conversion lookup table relating VLFF to the steatosis grade.
The output of the HepaFat-AI Analysis System is an automated report. This report is populated with the information stored in the DICOM header of the MRI images, and the analysis result Alpha converted into a VLFF value and a steatosis grade, associated confidence interval and normal range. The report also contains pictures of the 3 TEs of the analysed slice. This is essential for the radiologist to check if the image analysed is a liver image, and the result provided is consistent with other relevant clinical results.
Here's a summary of the acceptance criteria and study details for HepaFat-AI, based on the provided text:
Acceptance Criteria and Device Performance
| Acceptance Criteria (Steatosis Boundary / VLFF Threshold) | Device Performance (HepaFat-AI) - Sensitivity (95% CI) | Device Performance (HepaFat-AI) - Specificity (95% CI) |
|---|---|---|
| Grade 0 vs Grades 1-3 (4.1% VLFF) | 97.6% (93.3% to 99.2%) | 88.2% (65.7% to 96.7%) |
| Grades 0 & 1 vs Grades 2 & 3 (12.1% VLFF) | 86.1% (76.8% to 92.0%) | 74.8% (66.6% to 81.5%) |
| Grades 0-2 vs Grade 3 (16.2% VLFF) | 100.0% (81.6% to 100.0%) | 71.4% (62.4% to 78.1%) |
Note: The document states that "HepaFat-AI has a very similar diagnostic ability compared to HepaFat-Scan across all thresholds when both methods are tested directly against biopsy grades of steatosis. In particular, at the clinically important boundary separating no steatosis from any steatosis (Grade 0 vs Grades 1-3), the sensitivities and specificities of HepaFat-AI are no worse than HepaFat-Scan." The acceptance criteria were implicitly met by demonstrating comparable performance to the predicate device (HepaFat-Scan), which was previously cleared.
Study Details
-
Sample Size used for the test set and data provenance:
- Sample Size: 145 adult and pediatric NAFLD/NASH patients.
- Data Provenance: Not explicitly stated (e.g., country of origin), but implies clinical data from "two independent studies." The document doesn't specify if it's retrospective or prospective for the specific test set as an independent sample, but mentions it was used for "Clinical Validation."
-
Number of experts used to establish the ground truth for the test set and their qualifications:
- Not specified within the provided text. The ground truth was established via "liver biopsy with histological scoring of steatosis," but the number and qualifications of the pathologists performing the scoring are not detailed.
-
Adjudication method for the test set:
- Not specified. The ground truth was based on "histological scoring of steatosis," implying a single interpretation per biopsy, though it doesn't detail any multi-reader adjudication process for the biopsies themselves.
-
If a multi-reader multi-case (MRMC) comparative effectiveness study was done, and the effect size of how much human readers improve with AI vs without AI assistance:
- No MRMC comparative effectiveness study involving human readers with and without AI assistance is described. The study compared the standalone performance of HepaFat-AI to HepaFat-Scan against biopsy ground truth.
-
If a standalone (i.e., algorithm only without human-in-the-loop performance) was done:
- Yes, a standalone performance study was done. The reported sensitivities and specificities are for the HepaFat-AI algorithm itself, compared directly to biopsy results.
-
The type of ground truth used:
- Pathology (liver biopsy with histological scoring of steatosis).
-
The sample size for the training set:
- Not specified in the provided text. The document states, "Following the training of the AI tool, the system is 'locked-down' for final validation prior to release in commercial use," but does not give the number of samples used for this training.
-
How the ground truth for the training set was established:
- Not specified in the provided text. It can be inferred that the ground truth for training would also involve histology, given the ground truth for the validation set, but the method for establishing it on the training set is not detailed.
§ 892.1000 Magnetic resonance diagnostic device.
(a)
Identification. A magnetic resonance diagnostic device is intended for general diagnostic use to present images which reflect the spatial distribution and/or magnetic resonance spectra which reflect frequency and distribution of nuclei exhibiting nuclear magnetic resonance. Other physical parameters derived from the images and/or spectra may also be produced. The device includes hydrogen-1 (proton) imaging, sodium-23 imaging, hydrogen-1 spectroscopy, phosphorus-31 spectroscopy, and chemical shift imaging (preserving simultaneous frequency and spatial information).(b)
Classification. Class II (special controls). A magnetic resonance imaging disposable kit intended for use with a magnetic resonance diagnostic device only is exempt from the premarket notification procedures in subpart E of part 807 of this chapter subject to the limitations in § 892.9.