Previous Abstract | Next Abstract
Printable Version
October 15, 2013
8:00:00 AM - 9:00:00 AM
Room Room 104-Area C
Computer Vision Assisted Analysis of X-Ray for Rapid Detection of Retained Surgical Foreign Objects
Vicko Gluncic, M.D.,Ph.D., Serge Kobsa, Ph.D., Mario Moric, M.S., Greg Shakhnarovich, Ph.D., Sameer Ansari, M.D.,Ph.D.
RUSH, Chicago, Illinois, United States
Introduction: Each year, between 4,500 and 6,000 patients leave the nation's operating rooms with unintentionally retained surgical foreign objects (RSFO) in their bodies; with surgical sponges and needles accounting for more than two-thirds of all such incidents. In addition to serious complications, hospitalizations involving RSFOs increase health care costs per patient more than $60,000 on the average, while malpractice suits average $150,000 per case. Current recommendations for prevention of RSFOs include methodical wound exploration, usage of standardized practices for surgical items counting, usage of items with radio-opaque markers, and operative field X-rays (XR) before wound closure when item count discrepancy occurs. Radiographic screening is also recommended at the end of an emergent surgical procedure and in the cases of unexpected change in the procedure. Therefore portable XR protocols are crucial for RSFO detection. The problems with their use are the relatively low sensitivity (<60%) of identification of the human operator and the fact that radiologists and surgeons do not take formal training for the detection of RSFOs. In addition, these protocols require significant time for completion and final read. Since computer vision is superior to the human eye in the detection of defined objects, we developed software capable of rapid detection of the RSFOs in XRs, intended to be integrated into user interfaces of portable XR machines.

Methods: Software prototype based on pattern recognition algorithms was coded in MATLAB (MathWorks™, Natick, MA) and capable of identifying Accu-Sorb X-Ray Detectable USP Type VII Gauze with radiopaque marker (Medline Industries Inc., Beijing, China) (GAZ) and 2-0 Metric, 3/8, 24mm cutting needle (Syneture Covidien, Mansfield, MA, USA) (NDL) from XR images in any given 3D projection. XR images of the GAZ (N=30) and NDL (N=30) at various scales, exposures, and 3D rotations were acquired. Chest XRs (CXRs), cleared of patient information, were subsequently used to overlay these images, thus simulating GAZ and NDL appearance on regular CXR; i.e., GAZ on CXR (N=30), NDL on CXR (N=30), and CXR only (N=20). The software was evaluated with three sets of images; against the original set of GAZ and NDL XR images, against the overlaid images, and against the controls - plain CXR with no RSFO.

Results: In the first set, the software prototype showed 100% sensitivity and 100% specificity for both RSFOs. The second set (overlay set) showed 93% sensitivity, 100% specificity for NDL and 92% sensitivity, 100% specificity, for GAZ. There were no false positive detections of the RSFO.

Discussion: Costs of RSFO related sentinel events justify the investment in the development of this technology that has potential to greatly improve the efficacy of RSFO detection (higher sensitivity and specificity, faster/instantaneous detection, less expensive) and patient safety.

References: Gluncic V. International Patent Application No.: PCT/US2012/067070. Systems and methods for identification of implanted medical devices and/or detection of retained surgical foreign object from medical images. PCT US Receiving Office 2012

Copyright © 2013 American Society of Anesthesiologists