**5. Enhancement of laparoscopic and minimally invasive surgery**

In addition to aiding in tumor resections, computer vision is likely to impact many other aspects of surgery, especially with the increased integration of minimally invasive and robotic surgery [70]. Computer vision ML algorithms in the future may be able to process real time the videos taken during minimally invasive surgery (MIS) and robotic surgery, providing the surgeon with a broad array of additional, structured, and potentially actionable information. For example, computer vision algorithms may be useful in enhancing laparoscopic images. Given the anatomy of the abdomen, one issue common to an entire range of laparoscopic video signals is the quality of images. Nonuniform lighting, light-absorbing surfaces and substances (e.g., blood), along with other reasons for low endoscopic visibility, may lead to increased surgical risk and decreased efficiency in the operating room (OR) [71]. Because of these potential setbacks, computer vision algorithms may be able to process laparoscopic images in real time, digitally increasing lighting, removing vapor haze, and potentially filling in aspects of the image that may be obscured due to low visibility [72]. These applications have the potential to greatly improve ease-of-use of laparoscopes during surgery, decreasing the risk of incorrect targeting and decreasing the amount of time spent operating.

Further integration of computer vision in surgery could even lead to better identification of important anatomical landmarks in minimally invasive and robotic surgery. As mentioned previously, computer vision has already been used to identify objects in images and faces in security videos, and a logical extension of these uses would be the capacity to identify important surgical landmarks. For instance, rates of bile duct injury in laparoscopic cholecystectomies (LCs) have been seen to hover

around 0.45–0.8% [73, 74]. One of the most common causes of bile duct injury in LCs is misidentification of the common bile duct for the cystic duct [75]. An ML model trained on imaging data from laparoscopic surgeries was developed to identify critical anatomy in LCs in video with near-human accuracy, potentially leading to reduced risk of bile duct injury in LC in the future [76]. The largest challenge in building a model for this use would be the requirement for labeled video information. More specifically, any actionable model would need to be trained on many videos of laparoscopic surgeries in which the cystic duct is pre-identified in each of the thousands of frames within each training video. This formidable task is further complicated by the natural anatomical variations in human anatomy, necessitating the need for an even larger test data set of "normal variants" that can be encountered in the OR. Despite current limitations, it is likely only a matter of time before high fidelity models can be created, with significant resultant downstream benefits.

Of importance, AI/ML may also play a role as a component of augmented reality (AR) in surgery [77, 78]. One example with relatively mature application of AR is the area of spine surgeries, such as using the XVision Spine System (Augmedics, Arlington Heights, IL, USA) [79]. In this instance, AR-guided surgery works by using CT or MRI imaging to develop a three-dimensional (3D) model, then employing the AR program to overlay the model on the patient using AR glasses or other image projection modalities. Though this is a relatively new technology, initial studies investigating the use of AR systems in cadaveric pedicle screw placement indicate an absolute increase of accuracy from 88% (via fluoroscopy) to 94% (via AR guidance) [80]. In the immediate future, AR implementations will most likely be concentrated in orthopedic surgery and neurosurgery due to the relative immobility of bones and the spine compared to visceral organs. However, the potential increased use of peri- and intraoperative imaging in abdominal and thoracic surgeries may increase the viability of AR guidance in other operation theaters [81, 82].
