“Upon using a series of cameras and microphones in an OR during elective laparoscopic surgery, variations in surgeons technical skills, environmental distractions and intraoperative errors were noted. Taken together, these results demonstrate that the presence of cameras may be useful in identifying areas for improvement related to patient care and safety relevant to anesthesiologists.”
OR cameras have been used for a wide range of tasks ranging from clinical diagnosis to quality improvement to resident training. For example, by using one minute of endoscopic video recorded during a carotid artery injury in a high-fidelity cadaveric simulator, it is possible for a deep neural network to accurately predict the future blood loss (Sci Rep 2022;12:8137). Another study analyzed laparoscopic images to predict the amount of procedure time remaining in a cholecystectomy, which could be useful for optimizing OR workflow (IEEE Trans Med Imaging 2019;38:1069-78). Upon using a series of cameras and microphones in an OR during elective laparoscopic surgery, variations in surgeon’s technical skills, environmental distractions, and intraoperative errors were noted (Ann Surg 2020;271:122-7). Taken together, these results demonstrate that the presence of cameras may be useful in identifying areas for improvement related to patient care and safety relevant to anesthesiologists.
However, the use of OR cameras is not without controversy. Concerns about data, privacy, and litigation abound. One such pressing question is who owns the data? In one study, 88% of patients surveyed about OR video recordings believed the video belonged to them (Ann Surg 2022;276:e1057-63). However, there are numerous people involved in surgical care who might also be identifiable in the video. Recent developments in anonymization tools may help mitigate some privacy concerns, but consent requirements, ownership, and permittable usage are areas of ongoing debate (Int J CARS 2023; Surg Endosc 2022;36:3772-4). Although hospital risk management may be concerned about use of video recordings in a lawsuit, multiple legal cases demonstrate that video data is more likely to exonerate health care workers than implicate them (Surg Endosc 2022;36:3772-4). In fact, following a series of scandals, South Korea now requires cameras in ORs (asamonitor.pub/3CmGSVo).
There are clearly many uses for cameras mounted throughout the OR. However, for detection of small objects or recording kinetic activities performed by individuals, body cameras may be more advantageous (Plast Reconstr Surg-Glob Open 2022;10:e4315). At the University of Washington, through my FAER Mentored Research Training Grant (MRTG), we are using head-mounted cameras to identify drugs as they are being given to patients to improve medical record accuracy and minimize electronic recordkeeping burden. This may also improve patient safety as 20% of medication administration errors are due to incorrect dosing in anesthesia practice, with much higher rates reported for pediatric patients where weight-based dosing is more common and margins of error can be much smaller (Anaesth Intensive Care 2001;29:494-500; Br J Anaesth 2018;120:563-70; J Clin Anesth 2018;49:107-11). Head-mounted cameras can be preferable in such situations as there are potentially multiple sites for medication injection and so the camera can move with the user rather than focusing on a single injection site. Additionally, syringes and text labels are quite small, so the camera must be relatively close to obtain adequate image resolution. Lastly, a camera close to the eyes ensures that the medication label is simultaneously visible to the camera when the provider checks the drug and that the plunger can be seen during administration, whereas a chest-mounted camera might be blocked by a surgical drape or at a poor angle for visualization.
To accurately determine the drug dose during delivery, the name, concentration, and volume of the drug must be quickly ascertained. Prior work has been done to determine drug volume, but this was performed using colored medications and a stable, stationary location, which is far from the norm in the OR (J Pharm Innov 2019;14:341-58). My team and I at the University of Washington aim to identify the volume of clear liquids in a variety of commonly used syringe sizes through computer vision and deep neural network techniques. Early results from a series of laboratory-collected images aim to distinguish between the various syringe parts and the background. A series of images of syringes that was identified with our algorithm is shown in the accompanying figure. The plunger is in blue, drug in green, and syringe body containing air in red.
Our algorithm correctly classified each pixel in the image to the correct group (syringe, plunger, background, etc.) with an accuracy of 97% and mean absolute error in syringe volume determination of 0.12 mL. Now we are applying our algorithm to a data set including hundreds of real patient drug delivery events to assess performance with the aim of eventual clinical adaptation. More work is needed, both through my FAER MRTG and in the broader research community, before we understand the full potential for cameras to augment clinical practice in the perioperative space. I would encourage my fellow early-career physician investigators in anesthesiology to consider applying for a FAER grant to help them pursue this road of inquiry. There are clearly many potential uses for cameras beyond smartphones to improve patient care in the OR, and further research in this area is warranted.