Vision-based Detection and Tracking of Surgical Instruments

Retained surgical items (RSIs) are surgical instruments, such as laparotomy sponges and suture needles, that are le‰ in a patient’s body post-surgery. Œese incredibly costly mistakes also pose a serious health risk. Manual methods of tracking are highly inaccurate at scale, and modern methods (including embedded RF chips) are costly and limited to larger objects. We implemented a vision-based tracking system using two mounted webcams that runs n the background and only alerts the user if an object is missing or if an object is in the incorrect area, as a proxy for quantifying contamination risk. We constrained our model to a set of €ve surgical tools: laparotomy sponge, bovie tip, needle driver, surgical scissors and a needle counter box. We assembled and pre-processed our own dataset of 4650 images of these surgical items, and trained a Faster RCNN model on the data. Our resulting [email protected] was .6109. We have identi€ed vision-based surgical instrument tracking as a cost-e‚ective and accurate alternative to traditional methods.