Multiclass Object Detection for Surgical Instruments

Currently nurses prepare a surgical tray using a checklist for the instruments or some of the experienced nurses have it memorized. This leaves a lot of room for human error. A study showed that there are 1-5 cases of a near miss sharp (NMS) where a lost sharp (needle, blade, instrument, guidewire, metal fragment) is recovered prior to the patient leaving the operating room. An average of 21-30 min is spent managing each NMS, making a lost sharp event result in up to 70 min of added OR time.

To assist in the OR, in this paper I suggest a vertically down looking camera that clicks pictures of the tray and detects all the instruments present using the YOLOv3 model for object detection using bounding boxes. I believe having an account for all the instruments at all times will not only put the nurse’s minds at ease but also decrease the need for managing NMS.

To learn more: