AI-Rad Companion (Pulmonary)

The AI-Rad Companion (Pulmonary) is an advanced image processing software intended to provide quantitative and qualitative analysis of adult lung CT scans, supporting clinicians in the evaluation and assessment of pulmonary diseases. It leverages deep learning algorithms for precise segmentation of lungs, lobes, and lesions, and assesses lung parenchyma by identifying areas with low Hounsfield values from DICOM images produced by multiple CT vendor systems.

The AI-Rad Companion (Pulmonary) is a software-only image processing application designed to assist radiologists and physicians in evaluating and assessing lung diseases in adult patients. It utilizes previously acquired Computed Tomography (CT) DICOM images to provide quantitative and qualitative analysis, with functionalities including the segmentation and measurement of the complete lung and individual lung lobes, identification of lung areas with abnormally low Hounsfield unit values (compared to a predefined threshold), and the segmentation, measurement, and lobar attribution of detected lung lesions. The system also interfaces with the external syngo.CT Lung CAD medical device to aid in the detection of solid pulmonary nodules and has been validated for data from Siemens Healthineers, GE Healthcare, and Philips CT systems using specific reconstruction methods. Technologically, AI-Rad Companion (Pulmonary) employs deep learning algorithms, notably for its enhanced lung lobe segmentation, which differentiates it from prior model-based approaches. The software processes CT data to deliver detailed anatomical and pathological insights, such as calculating the percentage of low attenuation volume using a fixed -950 HU threshold for lung parenchyma evaluation. Performance validation, conducted through non-clinical testing and a significant clinical data-based software validation study involving over 4,500 CT datasets, demonstrated high accuracy in lung lobe segmentation. This validation yielded superior DICE coefficients (ranging 0.95-0.98), mean surface distances (0.5-1.0 mm), and volume errors (1.5-3.5%) compared to its predicate device for segmentation tasks, with consistent performance across various subgroups.

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 deep learning algorithm for lung lobe segmentation vary across different CT scanner manufacturers and reconstruction kernels, beyond the validated Siemens, GE, and Philips filtered backprojection/iterative reconstruction methods?; What is the clinical impact of using a fixed Hounsfield unit threshold (-950 HU) for parenchyma evaluation compared to a configurable threshold, particularly in diverse adult patient populations or varying disease severities?; Could the AI-Rad Companion (Pulmonary) be adapted or extended to analyze CT scans for interstitial lung diseases or other diffuse pulmonary conditions beyond nodule detection and basic parenchyma evaluation?; What are the specific failure modes of the lesion segmentation algorithm, and how does the system alert users to potentially inaccurate or incomplete segmentations of lung lesions?; How does the integration with syngo.CT Lung CAD enhance the standalone capabilities of AI-Rad Companion (Pulmonary) for solid pulmonary nodule detection and characterization, and what is the precise data flow and interaction protocol between these two systems?