AccuContour

AccuContour™ is a medical imaging software designed to assist radiation oncology departments by automatically registering multimodality images and segmenting non-contrast CT images for treatment planning, evaluation, and adaptation. The system utilizes deep learning algorithms for organ-at-risk contouring and intensity-based methods for image registration, with non-clinical tests demonstrating non-inferior performance to a predicate device based on DICE and NMI metrics.

AccuContour™ is a standalone medical imaging software developed by Xiamen Manteia Technology LTD. for use in radiation oncology departments. Its primary functions are to register multimodality images and segment non-contrast CT images, providing essential information for cancer treatment planning, evaluation, and adaptation. The software incorporates several key features: deep learning-based automatic contouring for organs-at-risk in the head and neck, thorax, abdomen, and pelvis; an intensity-based algorithm for automatic image registration across modalities such as CT/CT, CT/MR, and CT/PET; and manual contouring capabilities. Additionally, AccuContour™ supports standard medical imaging workflows, including receiving, editing, transmitting, importing/exporting medical images and DICOM data, patient management, and review of processed images. Non-clinical testing was conducted to evaluate the performance of AccuContour™ against a predicate device, MIM-MRT Dosimetry. Segmentation accuracy was assessed using the DICE similarity coefficient (DSC) on non-contrast CT images from both US and Chinese healthcare institutions, with ground truth established by consensus from multiple licensed physicians; the results indicated non-inferiority. Registration accuracy was evaluated using Normalized Mutual Information (NMI) on multimodality image sets, also demonstrating non-inferiority compared to the predicate. These tests covered images from major scanner vendors (GE, Siemens, Philips) and confirmed the device's compatibility with DICOM 3.0 compliant systems. No clinical studies were performed as part of this evaluation.

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 for segmentation in AccuContour™ account for anatomical variability across diverse patient populations beyond those represented in the US and China datasets?; What specific organs-at-risk within the head and neck, thorax, abdomen, and pelvis demonstrate the highest and lowest segmentation accuracy (DICE scores) with the deep learning model, and are there known limitations for particularly challenging anatomical structures?; To what extent is the performance of the intensity-based registration algorithm affected by variations in patient positioning or significant differences in image acquisition protocols across modalities like CT, MRI, and PET?; Beyond the non-inferiority demonstrated by DICE and NMI metrics, what is the practical clinical impact of any performance differences observed between AccuContour™ and the predicate device, especially in complex oncological cases?; How does AccuContour™ manage images with substantial artifacts or suboptimal quality, and what is the resulting impact on the reliability of its deep learning segmentation and intensity-based registration functionalities?