Feasibility and Utility of Artificial Intelligence (AI) / Machine Learning (ML) - Driven Advanced Intraoperative Visualization and Identification of Critical Anatomic Structures and Procedural Phases in Laparoscopic Cholecystectomy

Bile duct injury (BDI) during cholecystectomy is a serious surgical complication with increased risk of early death, serious ongoing morbidities including multiple reinterventions requiring prolonged and repeated hospital stay, and over a billion dollars in additional healthcare costs in US each year. The introduction laparoscopic approach has progressively increased overall laparoscopic cholecystectomy procedure volume, for clearly proven benefits of minimally invasive approach. However, despite the benefits, minimally invasive approach has resulted in increased and persistent incidence of BDI up to 4-folds in some reports.

Most major biliary injuries result from unrecognized or unintended perception, either misidentification or misinterpretation of the common bile or hepatic duct as the cystic duct or misidentification of an aberrant bile duct. It is increasingly clear that the routine use of intraoperative cholangiography (IOC) has a significant association with decreased and earlier intraoperative detection of BDI. The recent 'state of the art consensus conference on prevention of bile duct injury during cholecystectomy' in 2018 strongly recommended that documenting the critical view of safety (CVoS) and a liberal use of IOC in anticipated difficult cholecystectomy are highly recommended steps that can potentially mitigate the risk of BDI during laparoscopic cholecystectomy (LC) in patients especially with uncertain anatomy or difficult dissection.

Performing traditional IOC laparoscopically under fluoroscopic guidance however can challenge surgeons' skills, is time-consuming, and requires a learning curve to interpret images. Despite well-known and accepted risk factors for difficult cholecystectomies and potential for BDI, the use of advanced imaging for IOC however remains variable and highly user dependent. The rationale for selective use of IOC is attributable to the fact that the prevalence and incidence of BDI are low that an individual practitioner would infrequently encounter such complication in their daily practice or lifetime of a surgeon to justify a routine use. This is further compounded by the fact that radiation-based fluoroscopic IOC are cumbersome to efficient workflow, utility and cost (need for contrast reagent preparation and injection, potential adverse allergic reactions, risk of radiation exposure, training requirement, need for large capital equipment, space and cost), and variability in proficient analysis.

Recent randomized controlled trials using near-infrared fluorescent cholangiography (NIFC) using indocyanine green (ICG) demonstrated significantly superior visualization of extrahepatic biliary structures during laparoscopic cholecystectomy to white light (WLI) alone. Pre-dissection surgeon detection rates on naked eye were significantly higher (> 1.8 - 3.1 folds) with NIFC use for all 7 biliary structures than traditional WLI alone. However, although similar intergroup differences were observed for all structures, addition of NIFC did not improve additional detection of cystic duct and cystic duct/gallbladder junction after dissection has been done. In addition, increased body mass index was associated with reduced detection of most structures in both groups, especially before dissection. Interestingly, only 2 patients, both in the WLI group, sustained a biliary duct injury.

ActivSightTM is an FDA-cleared device that combines ICG fluorescence for extrahepatic biliary visualization and laser speckle contrast imaging (LSCI) for perfusion detection in a laparoscopic form factor. ActivSightTM allows augmented visualization to any current WLI laparoscopic visualization system displaying both extrahepatic biliary ICG and microperfusion over cystic duct and artery. As a non-significant risk device, ActivSightTM has been used in well over 150 patients for laparoscopic cholecystectomies and bariatric, esophageal, and colorectal procedures, with proven safety and utility. Moreover, ActivSightTM allows raw infrared visualization data for advanced analysis and AI/ML model development.

Surgeons/scientists now have segmented and analyzed different procedural phases of LC, and developed inferencing models and algorithms using artificial intelligence (AI)/machine learning (ML) based on standard WLI procedural videos. Although AI/ML models recognizing different phases of LC procedures can be as accurate as 80-95% on limited trained dataset once structures have been clearly dissected, more relevant and key value would be in identifying critical structures such as bile ducts and arteries before and during, not after surgical dissection has been performed. Early detection and identification of these critical structures before and during dissection of the triangle based on AI/ML trained on computer vision may aid surgeons in performing more effective and safer LC.

ActivSightTM is an FDA 510(k)-cleared optical imaging system based on monochromatic coherent light known as Laser-speckle-contrast imaging (LSCI) and represents a label-free imaging method using coherent monochromatic light where blood flow and tissue perfusion can be detected. A small imaging module that fits between any existing laparoscope and camera systems and a separate light source placed along any current commercial system will deliver objective real-time tissue perfusion and blood flow information intraoperatively. In addition, ActivSightTM can also effectively display with push of button an ICG-based visualization of the biliary tree in real time at equivalent or superior to current commercial products. The innovative form factor of ActivSight enables any laparoscopic system for ICG-based visualization at a fraction of the cost of current competitor with minimal disruption to workflow.

ActivInsightTM is a prototype software feature in ActivSightTM that recognizes procedural phases and critical anatomic structures, namely gallbladder, cystic duct, and cystic artery during LC using AI/ML-based algorithms. The key difference of ActivInsightTM to other models reported in the literature is that ActivInsightTM is trained on dataset annotated using ICG and LSCI in addition to WLI.

In this trial, the investigators propose to first validate and then test the precision and accuracy of ActivInsightTM in detecting critical phases and structures and compare the performance of the algorithms to those based on models developed using WLI only and traditional surgeon's naked eye detection during LC in pre-and mid-dissection phases. To eliminate any bias, the investigators will connect a secondary screen blinded to the main operating surgeon screens so that real-time function of the AI/ML is not visible to operating surgeons, to perform real-time analysis and comparison of routine use of advanced augmented visualization versus current WLI visualization alone with or without computer vision-based AI/ML.