QOCA® image Smart RT Contouring System

The QOCA® image Smart RT Contouring System is a software tool utilizing deep learning algorithms to automatically segment organs at risk from DICOM CT images for radiation therapy planning in adult patients. Its methodology relies on processing CT data to produce RTSTRUCT files, which then require clinical review and validation within existing treatment planning systems to ensure accurate OAR delineation in the head, neck, and pelvis regions.

The QOCA® image Smart RT Contouring System is a post-processing software designed to assist in radiation therapy planning by automatically contouring anatomical structures, specifically organs at risk (OAR), from DICOM CT images of adult patients. It focuses on the head, neck, and pelvis regions, employing deep-learning-based algorithms to process CT data (both contrast and non-contrast enhanced) and generate RTSTRUCT objects as output. These outputs are intended for use by radiation oncology departments as input for external beam radiation therapy treatment planning. A key aspect of its methodology is that it requires human oversight; generated contours must be reviewed, edited, and validated using separate Treatment Planning Systems or Interactive Contouring applications. The system does not provide its own visualization tools, relying on a web-based interface for settings and status management, and is not designed for lesion detection. The system's contouring accuracy was evaluated through nonclinical testing using a combination of private datasets from Taiwan and public datasets from the United States, encompassing images from various CT scanner manufacturers including GE, Siemens, Philips, and Toshiba. For initial model development, over 650 CT cases (317 head and neck, 351 pelvic) were used for training, validation, and testing. A dedicated segmentation performance study utilized an independent set of 220 CT images (110 head & neck, 110 pelvis) from Taiwanese and US public sources, with ground truth annotations based on established consensus guidelines. Performance metrics, primarily the Dice Similarity Coefficient (DSC) and Hausdorff Distance 95th percentile (HD95), demonstrated that the device achieved a median DSC greater than 0.80 for most evaluated OARs, such as the brain stem (DSC 0.942), mandible (DSC 0.956), and bladder (DSC 0.959). The system's performance was consistent across both contrast and non-contrast CT images, and organs not meeting predefined accuracy criteria were excluded from its scope of use.

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 deep learning model's performance vary across different CT scanner manufacturers and imaging protocols beyond those explicitly listed in the validation studies?; What are the specific predefined acceptance criteria (e.g., DSC thresholds) for OARs that led to some organs being excluded from the product's intended use, and which organs were specifically excluded?; What is the typical time savings or efficiency gain observed when using the QOCA® system for contouring compared to fully manual contouring in a clinical workflow, based on user studies or simulations?; How does the system handle anatomical variations due to pathology, prior surgery, or significant weight changes in the head, neck, and pelvis regions, and what is the impact on contouring accuracy in such cases?; Given that system settings and progress are managed via a web-based interface, what data security and integrity protocols are in place for the transfer and storage of DICOM data and RTSTRUCT objects?