Rayvolve

Rayvolve is a standalone software utilizing supervised deep learning algorithms, trained on a diverse international dataset of osteoarticular radiographs, to assist clinicians in the detection and localization of fractures in adult patients. Its methodology involves processing DICOM X-ray images to provide diagnostic aid, aiming to enhance fracture detection accuracy as demonstrated by improved reader performance in clinical studies.

Rayvolve is a standalone software medical device designed to assist radiologists and emergency physicians in detecting and localizing fractures on osteoarticular X-rays for adult patients aged 22 years and older. It employs deep learning techniques, developed using a diverse dataset of labeled radiographs from multiple international centers, to analyze DICOM images from various anatomical regions including the ankle, clavicle, elbow, forearm, hip, humerus, knee, pelvis, shoulder, tibia/fibula, wrist, hand, and foot across frontal, lateral, oblique, and axillary views. The software integrates with Picture Archiving and Communication System (PACS) servers, processing DICOM files to output predictions on the potential presence or absence of fractures, and is intended as an adjunctive tool, not a replacement for medical professionals. Performance data for Rayvolve includes both standalone assessments and a multiple-reader, multiple-case (MRMC) clinical study. Standalone testing on 2626 radiographs demonstrated high sensitivity (0.98763), specificity (0.88558), and an Area Under the Curve (AUC) of 0.98607, with consistent performance across different anatomical areas, patient demographics, and X-ray machine manufacturers. The MRMC study, involving 24 clinical readers and 186 cases, showed that Rayvolve-aided readings significantly improved diagnostic accuracy (AUC increased from 0.84602 to 0.89327), sensitivity (from 0.86561 to 0.9554), and specificity (from 0.82645 to 0.83116) compared to unaided readings in detecting fractures.

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 Rayvolve's deep learning model architecture specifically address the variability in fracture presentation across different anatomical regions and X-ray views?; What were the specific criteria for labeling radiographs in the training dataset, and how was inter-rater reliability ensured among labellers from diverse international centers?; Beyond the listed AUC, sensitivity, and specificity, what is the software's performance on subtle or hairline fractures, which are often challenging to detect?; How does the software handle images with significant artifacts or poor image quality, and what is the false positive rate in such scenarios?; The MRMC study showed improved reader performance; what was the average time difference for radiologists to interpret cases with versus without Rayvolve assistance, and how does this impact clinical workflow?