HealthFLD

HealthFLD is an AI-powered image processing software designed to provide quantitative analysis of liver attenuation from CT scans, aiding clinicians in the assessment of fatty liver. The methodology involves deep-learning based liver segmentation and Hounsfield Unit measurement, with performance validated through high agreement with radiologist-established ground truth and specific thresholds for contrast and non-contrast studies.

HealthFLD is an image processing software designed to assist clinicians in the evaluation and assessment of Fatty Liver. It analyzes computed tomography (CT) images, both non-contrast and contrast-enhanced, to provide quantitative and qualitative measurements of liver attenuation, specifically the mean Hounsfield Units (HU) in a region of interest. The software utilizes a deep-learning algorithm for liver segmentation and automatically adjusts its display threshold for liver attenuation based on whether a portal venous contrast phase is detected (<40 HU for non-contrast/non-portal venous phase, <75 HU if portal venous phase identified). Key software modules include DICOM data input and validation, the core HealthFLD algorithm for analysis, an integration feature for results output to a hosting application, and an error reporting system. The HealthFLD software is indicated for patients aged 18 to 75 undergoing CT scans for any clinical indication, provided the scan includes a significant portion of the liver. It is important to note that HealthFLD is not intended to provide a definitive diagnosis or risk assessment of fatty liver disease but rather to support clinical evaluation. Performance validation was conducted via a retrospective study involving 250 CT scans, comparing HealthFLD's measurements against ground truth established by three board-certified radiologists. The software demonstrated high agreement with ground truth for liver attenuation thresholds (95.98% for <40HU vs ≥40HU; 98.39% for <50HU vs ≥50HU) and accurate identification of portal venous phase (95.98% agreement). Bland-Altman analysis showed acceptable limits of agreement for liver density bias ([-7.80HU, 7.00HU]), confirming its intended performance.

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 HealthFLD algorithm's performance in liver segmentation and HU measurement vary across different CT scanner models, manufacturers, and reconstruction kernels not explicitly detailed in the validation dataset?; What is the impact of varying levels of image noise or artifacts, commonly encountered in clinical CT scans, on the accuracy and reliability of HealthFLD's liver attenuation measurements?; Given the automatic adjustment of display thresholds based on portal venous phase detection (from <40 HU to <75 HU), what is the sensitivity and specificity of the algorithm in correctly identifying this phase, and how does misidentification impact the clinical utility of the attenuation measurement?; How does the software handle CT scans with partial liver coverage, and what is the minimum liver volume or specific anatomical landmarks required for reliable analysis?; Beyond mean HU, does HealthFLD provide any information on the heterogeneity of fat distribution within the liver, which could be clinically relevant for assessing patterns of steatosis?