CINA-CSpine

CINA-CSpine is a software application employing deep learning algorithms to analyze cervical spine CT images for suspected acute fractures, excluding vertebral compression fractures, by processing DICOM data and identifying patterns learned from a large training dataset. Its purpose is to enhance radiological workflow efficiency by providing timely notifications and prioritizing potentially critical cases on a worklist, thereby aiding trained physician specialists in the triage process for earlier patient intervention.

CINA-CSpine is a radiological computer-assisted triage and notification software designed to identify suspected acute cervical spine fractures in CT scans, thereby assisting trained physician specialists in prioritizing their review workload. The system, which runs on standard server hardware and can integrate with the CINA Platform or similar medical image communication systems, receives cervical spine CT images and processes them using deep learning algorithms. These algorithms analyze the images to detect findings compatible with acute fractures, such as non-displaced fracture lines or displaced fragments, and then generate notifications for these flagged studies to a worklist application, which also displays a compressed, non-diagnostic preview image for informational purposes, without altering the original medical image. The scientific basis of CINA-CSpine involves deep learning models extensively trained on 1,338 cervical spine CT cases sourced from the US and France, covering a range of fracture presentations, locations, and acquisition protocols from multiple scanner manufacturers; additional deep learning models are used to locate individual vertebrae and exclude images not conforming to the expected field of view. A retrospective, multicenter, multinational, blinded validation study on 328 anonymized non-contrast CT scans demonstrated the software's performance, achieving a global sensitivity of 90.3% and a specificity of 91.9% for detecting cervical spine fractures, benchmarked against a consensus of three US-board-certified expert radiologists. The system also showed efficient processing, with a mean time-to-notification of approximately 2.9 minutes for all cases, supporting its role in expediting the review of critical findings, although it is not designed for direct diagnosis or the detection of vertebral compression fractures, and clinicians remain responsible for reviewing full images per standard of care.

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 CINA-CSpine algorithm's performance vary with different types of cervical spine fractures (e.g., non-displaced vs. displaced, specific vertebral levels, or fracture severity)?; What is the impact of image quality variations (e.g., noise, artifacts, different reconstruction kernels, or slice thickness variations beyond the training set) on the CINA-CSpine's detection accuracy and notification reliability?; Given that the software is not designed to detect vertebral compression fractures, what are the practical implications for workflow integration in emergency departments or teleradiology settings where these fractures are also a common concern on cervical spine CTs?; How does the training dataset composition of 1,338 cases, particularly the distribution of various fracture types, patient demographics, and acquisition protocols, influence the generalizability and potential biases of the algorithm when applied to diverse, real-world patient populations and rare fracture patterns?; What mechanisms are in place for ongoing monitoring of CINA-CSpine's performance post-deployment in clinical settings, and how are algorithm updates or retraining handled to maintain or improve accuracy over time as imaging technologies evolve or new fracture patterns emerge?