iCAS-LV

The iCAS-LV is a software device designed to aid physicians in the evaluation of liver lesions from CT scans by providing tools for image processing, including AI-driven segmentation and volumetric analysis. Its methodology involves integrating with PACS, utilizing 3D U-Net models for lesion segmentation, and enabling quantitative assessment of lesion characteristics and changes over time to support clinical evaluation.

The iCAS-LV is a PC-based, noninvasive image analysis software application designed to assist trained physicians in evaluating physician-identified liver lesions from CT scans. It receives multi-phase volume datasets of reconstructed CT studies from PACS devices, processes them using a suite of tools for visualization, measurement, segmentation, annotation, image registration, and reporting, and then transfers the processed output back to PACS in DICOM format. The software's workflow involves generating liver regions of interest from prior and current scans, registering these ROIs using deformable registration, and employing 3D U-Net models for lesion segmentation, which radiologists then review and can manually correct. It also facilitates semi-automatic matching of lesions between scans and quantifies changes in lesion volume and diameter over time. The system utilizes a deep learning algorithm (3D U-Net models) for lesion segmentation and supports a pipeline that includes supervised automatic liver registration and semi-automatic lesion matching. Performance was evaluated through phantom studies and clinical data analysis. Phantom tests demonstrated that iCAS-LV's volume and diameter estimations for spheres were comparable or slightly better than those of expert radiologists across multiple CT scanner platforms (GE, Philips, Siemens, Canon). Standalone clinical performance testing, using a retrospective multi-site dataset of 108 patients and 2,127 liver lesions, showed that physician-assisted iCAS-LV assessments of 3D lesion volumetrics agreed with ground truth established by three radiologists, supporting its utility based on metrics like DICE, ASSD, SHD, volume, and RECIST measurements. The device is intended to support physicians but not for isolated diagnosis or patient management decisions.

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 performance of the 3D U-Net model for liver lesion segmentation vary with lesion morphology, size, and proximity to vascular structures or other organs?; What is the typical time saved by a radiologist when using iCAS-LV for a complex case with multiple liver lesions compared to manual assessment methods?; How does the semi-automatic lesion matching algorithm handle challenging cases, such as significant changes in liver morphology or lesion appearance between scans?; What specific training is recommended for physicians to achieve optimal proficiency and ensure accurate use of the iCAS-LV, particularly for manual correction of segmentations?; Are there plans to extend the iCAS-LV's capabilities to other organs or imaging modalities beyond CT for liver lesions?