AccuContour

AccuContour is a medical imaging software leveraging deep learning for organ-at-risk segmentation on non-contrast CT and intensity-based algorithms for multi-modality image registration, designed to support radiation oncology. Its purpose is to provide accurate anatomical delineation and image fusion, alongside tools for treatment plan evaluation and dose analysis, thereby facilitating more precise treatment planning, evaluation, and adaptation in radiotherapy.

AccuContour is a standalone medical imaging software designed for radiation oncology departments to aid in treatment planning, evaluation, and adaptation. Its core methodology involves the registration of multi-modality images (including CT, MRI, PET, and CBCT for deformable registration) and the segmentation of non-contrast CT images. The software utilizes deep learning algorithms for the automatic contouring of organs-at-risk across various anatomical regions such as the head and neck, thorax, abdomen, and pelvis, and employs intensity-based algorithms for both rigid and deformable image registration. Beyond image registration and segmentation (which includes manual contouring options), AccuContour offers a suite of general and advanced functions. These include comprehensive DICOM data and patient management, tools for reviewing processed images, and an extension tool. Significantly, it incorporates features for plan evaluation, such as the display of DICOM RT plans, isodose lines, DVH statistics, and RT plan comparison, alongside dose analysis capabilities like DICOM RT dose display and dose accumulation. Performance evaluation focused on its deformable registration accuracy, which was found to be non-inferior to a reference system based on Normalized Mutual Information metrics; software verification and validation were also completed.

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 algorithm's performance in contouring organs-at-risk compare across different anatomical regions (head/neck, thorax, abdomen, pelvis) and in diverse patient populations, particularly concerning edge cases or subtle anatomical variations?; What specific image features or intensity characteristics are primarily utilized by the "Intensity Based" algorithm for deformable registration, and how robust is it to variations in image quality, noise levels, or acquisition protocols across different CT, MRI, PET, and CBCT scanners?; What are the quantitative metrics used by AccuContour for "plan evaluation and plan comparison," and how does the "dose accumulation" feature account for potential uncertainties in deformable image registration when mapping doses across different image sets or treatment fractions?; What is the typical end-to-end processing time for a standard clinical case (e.g., head and neck) involving multi-modal registration, deep learning-based OAR contouring, and subsequent plan evaluation using AccuContour, and how does this integrate into existing clinical workflows?; The software validation considered it a "major" level of concern. What specific validation tests were performed to ensure the safety and accuracy of the deep learning segmentation and deformable registration components, especially concerning potential systematic errors or biases?