Search Results

Intended use:

HepaFatSmart is intended for the quantitative measurement of volumetric liver fat fraction (VLFF), proton density fat fraction (PDFF) and steatosis grading.

HepaFatSmart is an application that is used for the non-invasive evaluation of liver tissue by utilising magnetic resonance images to evaluate the difference in resonance frequencies between hydrogen nuclei in water and triglyceride fat. The quantitative triglyceride fat fraction is based on the measurement of a magnetic resonance parameter that reflects the ratio of the proton density signal of triglyceride fat to the total proton density signal in the liver.

Indications for use:

Support clinical diagnoses in individuals with confirmed or suspected fatty liver disease;

Support the subsequent clinical decision making processes for patients under management for fatty liver related disease or metabolic syndromes;

Aid in the assessment and screening of living donors for liver transplant.

Results, when interpreted by a trained physician can be used to support clinical diagnoses about the status of liver fat content, the subsequent clinical decision making processes for the management of fatty liver related diseases, metabolic syndromes, liver donor screening and lifestyle change. HepaFatSmart can be used to analyse the MRI images of patients of all populations independent of age and gender, with suspected clinical conditions related to the level of liver fat.

HepaFatSmart is an SaMD designed to automatically analyse magnetic resonance imaging (MRI) datasets for quantitative assessment of a patient's liver fat, in form of volumetric liver fat fraction (VLFF), proton density fat fraction (PDFF), and steatosis grade. It is an Al assisted, automated version of HepaFat-Scan (another SaMD of Resonance Health). To carry out an analysis, the user simply uploads DICOM images to FAST, Resonance Health's secured user portal and job management system. No other user input is required for the analysis thereby minimising the impact of human error on obtained results. HepaFatSmart requires DICOM images as input data that have been acquired according to the HepaFatSmart (same as HepaFat-Scan) protocol.

The provided documentation describes the acceptance criteria and a study proving that HepaFatSmart (V2.0.0) meets these criteria, demonstrating its substantial equivalence to the predicate device HepaFat-Scan and improved performance over HepaFat-AI.

Here's a breakdown of the requested information:

1. Table of Acceptance Criteria and Reported Device Performance

The document does not explicitly present a table of predetermined acceptance criteria for the new device prior to testing. Instead, it demonstrates the device's performance by comparing it to a reference standard (HepaFat-Scan) and an existing predicate (HepaFat-AI). The "acceptance" is implied by demonstrating substantial equivalence to HepaFat-Scan and improvement over HepaFat-AI, primarily through quantitative metrics like bias, repeatability, and limits of agreement, as well as sensitivity and specificity for clinically relevant thresholds.

However, based on the provided data, we can infer the performance metrics used for comparison, which implicitly serve as the targets for acceptance.

Inferred Acceptance Criteria (Implicitly compared to HepaFat-Scan and better than HepaFat-AI):

2. Sample Size Used for the Test Set and Data Provenance

• Test Set Sample Size:

  • Repeatability Study: n = 42 subjects initially; n = 41 subjects for HepaFatSmart analysis (one case identified as high iron and excluded by the new algorithm).
  • Validation Study: n = 300 initially; n = 281 datasets successfully passed the IQC rules and were used for analysis.

• Data Provenance: The document states the data used for the validation study comprised "fully quarantined 300 validation subjects with different clinical conditions across a broad age range and fat level scanned from different MRI centres with different MRI makes and models." While explicit countries of origin are not specified, the mention of "different MRI centres" and "different MRI makes and models" suggests a diverse, real-world dataset. The data appears to be retrospective, as it refers to a "quarantined dataset" used for validation.

3. Number of Experts Used to Establish the Ground Truth for the Test Set and Qualifications

The ground truth for this study is established by the reference standard HepaFat-Scan. HepaFat-Scan is described as a "Standalone software application to facilitate the import and visualization of multi-slice, gradient-echo MRI data sets... to provide objective and reproducible determination of the triglyceride fat fraction in magnetic resonance images of the liver." It performs quantitative triglyceride fat fraction measurements.

The document states that HepaFat-Scan is "Algorithmic, with human interaction for Region of Interest (ROI) selection." The "User" for HepaFat-Scan is listed as "Resonance Health's trained analyst."

Therefore:

• Number of Experts/Analysts: The document doesn't specify a number, but rather a type of user: "Resonance Health's trained analyst." This implies trained personnel, but not necessarily multiple independent experts in a consensus-building scenario.
• Qualifications of Experts: "Resonance Health's trained analyst." No further specific qualifications (e.g., years of experience, radiologist vs. technician) are provided for the individuals performing the HepaFat-Scan ground truth ROI selection. The emphasis is on the software being the reference standard.

4. Adjudication Method for the Test Set

The primary ground truth is derived from the HepaFat-Scan software with human ROI selection. The document does not describe any specific adjudication method (e.g., 2+1, 3+1 consensus by multiple readers) for establishing the ground truth measurements. The HepaFat-Scan output, with its human-selected ROI, appears to be accepted as the reference.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was Done

No, a traditional MRMC comparative effectiveness study was not explicitly done to show human readers improve with AI vs. without AI assistance.

This study is focused on the standalone performance of HepaFatSmart and its equivalence/superiority to existing automated or semi-automated software solutions (HepaFat-Scan and HepaFat-AI). HepaFatSmart is an "AI-assisted, automated version" of HepaFat-Scan. HepaFat-Scan itself involves "human interaction for Region of Interest (ROI) selection." The document implies that HepaFatSmart's full automation (no user input for analysis, AI-predicted ROI) minimizes human error and makes it potentially better or comparable to the human-assisted HepaFat-Scan.

The statement: "Bias and both repeatability coefficients for the HepaFatSmart are slightly better than those obtained from the repeated scans of HepaFat-Scan, indicating the performance of the HepaFatSmart is comparable (no worse) than human for the repeatability data analysed. This does not suggest yet that the HepaFatSmart is better than human analyst as the original human analysis (HepaFat-Scan) historically used two small liver ROIs rather than a single large liver ROI used in the HepaFatSmart with potentially slightly larger sampling error in the original HepaFat-Scan analysis."
And: "HepaFatSmart demonstrated 100% repeatable (reproducible) in the repeatability study using the same datasets analysed twice with zero VLFF difference between the first and second analyses, which is better than human analysis with HepaFat-Scan."
These statements imply a comparison to "human analysis" (as part of HepaFat-Scan's workflow), but it's not a formal MRMC study as typically understood where human readers' diagnostic performance is measured with and without AI assistance on the same cases.

6. If a Standalone (Algorithm Only Without Human-in-the-Loop Performance) was Done

Yes, a standalone performance study was foundational to this submission. HepaFatSmart is described as a "Standalone software platform" that "automatically analyse... MRI datasets... No other user input is required for the analysis thereby minimising the impact of human error on obtained results." The entire study evaluates the performance of this algorithm-only approach (HepaFatSmart) by comparing its output directly to the output of the human-assisted reference standard (HepaFat-Scan) and the previous AI version (HepaFat-AI).

7. The Type of Ground Truth Used

The ground truth used for the test set is expert-assisted software output (HepaFat-Scan).
Specifically, it's the "HepaFat-Scan" software's quantitative measurement of Volumetric Liver Fat Fraction (VLFF), which involves human interaction for Region of Interest (ROI) selection. This serves as the "reference standard" against which HepaFatSmart's automated results are compared.

8. The Sample Size for the Training Set

The document does not provide the sample size directly for the training set of the HepaFatSmart AI model. It notes that the system is "completely 'locked down' for final validation prior to release in commercial use to ensure reproducibility of the results" after training.

9. How the Ground Truth for the Training Set was Established

The document briefly mentions that HepaFatSmart uses "one (1) convolutional neural network (CNN) performing liver ROI detection... Following the training of the Al assisted device, the system is completely 'locked down' for final validation..."
Given that HepaFatSmart is an "AI-assisted, automated version of HepaFat-Scan," it is highly probable that the ground truth for training its CNN for ROI detection was established by using data annotated or measured previously by the HepaFat-Scan methodology, likely involving the "Resonance Health's trained analyst" for ROI selection. However, the specifics of this process (e.g., how many cases, who annotated, adjudication) are not detailed in the provided text.

Ask a specific question about this device