ClearRead CT

ClearRead CT is a software application intended to improve radiologists' detection of pulmonary nodules in chest CT scans of asymptomatic individuals by generating a vessel-suppressed secondary image series with CADe markers. The system utilizes advanced image processing and machine learning algorithms for segmentation, vessel suppression, and nodule feature analysis to identify and characterize suspicious regions, thereby enhancing detection accuracy and reducing interpretation time.

ClearRead CT™, developed by Riverain Technologies, LLC, is a dedicated post-processing software application designed to assist radiologists in the detection of pulmonary nodules during the review of chest CT examinations from asymptomatic individuals. It functions as an adjunctive tool, creating a secondary, vessel-suppressed lung CT series with Computer-Aided Detection (CADe) marks and associated region descriptors for up to five Regions of Interest (ROIs) per scan; it is not intended as a substitute for the original CT series. The system receives the original lung CT series and applies advanced image analysis and machine learning algorithms through sequential steps: volume preparation, body and air segmentation, volume normalization for consistent image appearance, airway and lung segmentation, and a critical vessel suppression technique that removes vascular structures, bronchial walls, and fissure lines while aiming to preserve nodular structures, including those attached to vasculature, thereby enhancing nodule detectability. The core science involves ROI generation from the vessel-suppressed volume, followed by feature extraction—computing measurements such as volume, maximum/minimum diameter, and average density for each ROI—and ROI classification, which includes rejecting calcified nodules and limiting marks to the top five ranked ROIs. ClearRead CT is optimized for detecting solid, sub-solid, and ground glass nodules primarily between 5mm and 30mm. Clinical validation via a multi-reader multi-case (MRMC) study demonstrated significant improvements in radiologist performance; the partial area under the Localization Receiver Operating Characteristic (LROC) curve for detecting actionable nodules increased from 0.6786 (unaided) to 0.7469 (aided with ClearRead CT), a statistically significant improvement (p=0.0418). Furthermore, the use of ClearRead CT resulted in a statistically significant 26% reduction in radiologists' interpretation time, averaging a decrease of 29 seconds per case. Standalone machine performance testing on clinical data showed a sensitivity of 89.5% for detecting cancers at a false positive rate of 0.7469 per case.

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 vessel suppression algorithm differentiate between vascular structures and nodules, particularly for nodules attached to or abutting vessels?; What was the inter-reader variability in nodule detection and characterization both with and without the ClearRead CT, and how did the software impact this variability beyond the LROC curve improvement?; The system marks up to 5 Regions of Interest (ROIs) per CT series; what is the rationale for this limit, and how does the system prioritize which ROIs to mark if more than five suspicious regions are detected?; Given the system is optimized for detecting solid, sub-solid, and ground glass nodules between 5mm and 30mm, what is its performance on nodules outside this size range or of different morphologies?; The clinical study data included CT scans with slice thicknesses of 2-3.0mm; how does the performance of ClearRead CT vary with different CT acquisition parameters, such as thinner slices or varying radiation doses and reconstruction kernels?