Al-Rad Companion (Pulmonary)

The AI-Rad Companion (Pulmonary) is an AI-powered software that analyzes adult CT DICOM images of the thorax, providing quantitative and qualitative assessments of lung structures and abnormalities. It employs deep learning algorithms for segmentation of lungs, lobes, lesions, and opacities, and measures features such as pulmonary density and bi-directional lesion diameters to support clinicians in evaluating lung diseases.

AI-Rad Companion (Pulmonary) is an advanced software-only medical system designed to assist clinicians in evaluating lung diseases by analyzing adult Computed Tomography (CT) DICOM images. Utilizing artificial intelligence (AI), machine learning, and deep learning, it performs post-processing tasks such as landmark detection, segmentation, and classification. Its core functionalities include segmenting and measuring the complete lung, individual lung lobes, and lung lesions; identifying regions with lower Hounsfield values indicative of certain lung conditions; and detecting areas with elevated Hounsfield values, distinguishing between different types of opacities. The software version VA13 enhances these capabilities with features like "Pulmonary Density" analysis for segmenting opacity regions and calculating their percentage relative to lung volumes, and "Bi-directional lesion diameter" for more comprehensive lesion measurement, supporting both cloud and on-premises (edge) deployment. The system provides quantitative and qualitative analytical results to support the clinical assessment of diseases of the thorax. For instance, the "Pulmonary Density" feature segments opacity regions and can assign results to ranges defined by established criteria (e.g., Bernheim et al.), while the "Bi-directional lesion diameter" feature adds the maximum orthogonal 2D diameter to existing lesion measurements. Performance data from non-clinical testing and clinical evaluation demonstrated its effectiveness: lung lobe segmentation achieved average DICE coefficients ranging from 0.94 to 0.96 on 250 datasets. Segmentation of opaque regions was validated on 150 datasets, with 93.0% of the percentage of opacity (PO) values falling within the 95% Limits of Agreement (LoA) when compared to human reads. The software has been validated for data from Siemens Healthineers, GE Healthcare, and Philips CT systems, delivering results in various image formats, including DICOM Structured Reports.

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 AI algorithm differentiate between various types of lung opacities (e.g., ground glass, consolidation) beyond just identifying elevated Hounsfield values?; What is the variability in performance (e.g., DICE coefficients, Limits of Agreement for opacity) when the software is applied to CT images acquired with different scanner parameters (e.g., slice thickness, radiation dose, reconstruction kernels) not explicitly covered in the validation?; How does the 'Pulmonary Density' feature's assignment to ranges, such as those defined by Bernheim et al., correlate with clinical severity or specific pathological diagnoses across diverse patient populations?; What are the specific computational requirements and average processing times for the AI-Rad Companion (Pulmonary) on both cloud and edge deployment models, and how might these impact real-time clinical workflow efficiency?; Beyond the listed validation with Siemens Healthineers, GE Healthcare, and Philips scanners, what is the robustness and generalizability of the AI algorithms when encountering images from other CT manufacturers or those with significant motion or beam-hardening artifacts?