Prospective Silent Evaluation of AI for Retained Foreign Object Detection

Summary

Retained foreign objects, such as surgical sponges, needles, or instruments, are rare but potentially serious complications of surgery. When there is concern that an object may have been left behind, an intraoperative radiograph may be obtained while the patient is still in the operating room. These radiographs can be difficult to interpret because of overlapping surgical equipment, patient positioning, and the need for rapid interpretation. This prospective observational study evaluates an artificial intelligence (AI) system designed to detect retained foreign objects on intraoperative radiographs. Eligible radiographs obtained during routine clinical care will be automatically processed by the AI system in real time for approximately one year. The AI results will not be shown to radiologists, surgeons, or other members of the clinical care team and will not affect patient care. The study will evaluate how accurately the AI system detects retained foreign objects and how reliably and quickly the system operates in a real-world clinical environment.

Detailed Description

This is a prospective, non-interventional, silent observational study evaluating the real-world performance and operational feasibility of an artificial intelligence system for detection of retained foreign objects (RFOs) on intraoperative radiographs. Consecutive eligible OR count radiographs obtained during routine clinical care will be identified using institutional imaging protocol labels and automatically routed from the clinical imaging environment to a secure local research inference system within the institutional firewall. The AI system will process eligible images in real time and generate research outputs, including detection probabilities, localization bounding boxes, and processing timestamps. The study is conducted in a silent fashion. AI-generated results will not be displayed to interpreting radiologists, surgeons, or other members of the clinical care team and will not be used for clinical decision-making or patient management. No additional imaging, procedure, intervention, or patient contact is required as part of the prospective study. Reference standards for evaluation of AI performance will be established using available clinical information, including finalized radiology reports, operative documentation, follow-up imaging when available, and expert adjudication of discordant cases when needed. The study will evaluate standalone AI diagnostic performance using measures including sensitivity, specificity, and area under the receiver operating characteristic curve. Additional measures will evaluate localization performance and operational feasibility, including system latency from image acquisition to availability of the AI output. Exploratory analyses may evaluate performance according to factors such as retained object type, surgical context, and image characteristics. The prospective evaluation is planned to include approximately 12 months of consecutive eligible examinations and is intended to characterize the performance of the AI system under real-world clinical workflow conditions without influencing clinical care.

Eligibility Criteria