aprevo® Digital Segmentation

The aprevo® Digital Segmentation software utilizes an AI-based algorithm to automatically segment spinal anatomy, primarily the lumbar region, from DICOM images, producing a 3D model for use by trained design personnel. Its purpose is to provide an efficient tool for creating these models, with non-clinical testing validating its segmentation accuracy through Intersection over Union scores and vertebral body labeling metrics.

The aprevo® Digital Segmentation is a software medical device designed for use by trained medical design personnel to perform digital image segmentation of the spine, with a primary focus on the lumbar anatomy. Its core methodology involves processing DICOM images as input. This process includes pre-processing steps to filter out soft tissue and accurately identify the spinal structure. The software then employs an artificial intelligence (AI)-based algorithm to automatically segment the identified spinal anatomy and subsequently render a 3D model of the spine as its output. The software's functionality enables users to import medical images and generate digital 3D models, specifically of the spine. While the software itself does not include integrated tools for users to directly review and edit the segmentation or the 3D models, its output files can be reviewed and utilized within broader workflows using third-party software. Non-clinical testing demonstrated the software's performance, with an Intersection over Union (IOU) score for segmentation exceeding an 80% acceptance criterion, and vertebral body labeling accuracy surpassing a 90% overall acceptance criterion. Sensitivity and specificity for this labeling also exceeded 80% each, validated using diverse datasets representing various populations and equipment to ensure clinical relevance.

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-based algorithm differentiate between various spinal pathologies (e.g., fractures, scoliosis, stenosis) during the segmentation process, and what is its accuracy for these specific conditions?; What are the specific characteristics of the "key cohorts" used in the validation datasets, and did performance vary significantly across these different demographic or pathological groups?; Given that the software outputs a 3D model without integrated editing tools, what are the typical third-party software solutions recommended or commonly used for subsequent review, modification, and application of these models in pre-surgical planning or device design?; What are the computational requirements (e.g., processing power, memory) for running the aprevo® Digital Segmentation software, and how does its processing time scale with the complexity or length of the spinal segment being analyzed?; How does the pre-processing step specifically filter soft tissue and identify the spine, and what is its robustness to variations in image quality or patient anatomy (e.g., obesity, severe degeneration)?