Improving precision of laparoscopic cholecystectomy: A review on the role of newer intraoperative imaging aids and artificial intelligence

Abstract

Bile duct injuries have been the main area of concern in laparoscopic cholecystectomy (LC). Adoption of strategies such as identification of important anatomical landmarks, critical view of safety (CVS) achievement, careful use of energy sources, timeout and bailout techniques in difficult cases have evolved to increase the safety of LC. However, bile duct injury rates are yet to equal to that of open cholecystectomy. Newer technological advances in endovision systems, like indocyanine green near infra-red fluorescence cholangiography, yellow enhancement and integration of artificial intelligence (AI) appear to be promising. Evolving evidence has suggested the incorporation of AI-driven systems may improve safety of LC such as “go” and “no-go” zone indication, CVS assessment, surgical phase recognition, predicting surgical difficulty, automated coaching and training. This narrative review deliberates upon the current evidence in literature related to use of newer imaging aids and AI in LC.

Key Words: Laparoscopic cholecystectomy; Indocyanine green near infrared cholangiography; Yellow enhancement; Artificial intelligence; Critical view of safety; R4U line; Culture of safety cholecystectomy; Anatomic segmentation; Semantic segmentation; Instance segmentation; Go/No-Go zones

Core Tip: Endovision technology coupled with multispectral imaging, such as indocyanine green near infra-red, yellow enhancement, and artificial intelligence (AI), has potential to improve the precision of laparoscopic cholecystectomy by enhancing tissue differentiation. AI architectures can predict difficult cholecystectomy, assess surgical workflow, achieve anatomic segmentation, identify safe dissection zones, and assist in coaching/training. By leveraging multispectral imaging along with advanced AI architecture-convolutional neural networks for segmentation, graph neural networks for relational anatomical mapping, and transformers for temporal workflow analytics-surgical platforms can achieve context-aware surgery, ultimately bridging the gap between intraoperative imaging and enhanced patient safety through automated surgical coaching and decision support.

INTRODUCTION

Minimizing bile duct injury (BDI) remains the primordial crux of laparoscopic cholecystectomy (LC). Most cases of BDI stem from misidentification of anatomy rather than a lack of skill or dexterity. Adherence to a cognitive map using anatomical landmarks, coupled with the judicious use of adjunct imaging systems for achievement of the critical view of safety (CVS), along with implementation of time-out practices and consideration of bailout methods, are practices advocated to mitigate BDI in LC[1-4]. Although these measures have reduced the incidence of BDI, they have not yet achieved the safety metrics seen in the era of open cholecystectomy. Advances in endoscopic vision systems and the incorporation of artificial intelligence (AI)-aided hepatobiliary anatomy mapping may serve as valuable additions to the surgical armamentarium. This review highlights recent advances in this field, particularly in the context of evolving imaging systems and AI in enhancing the safety and precision of LC.

CVS, CULTURE OF SAFETY, AND ANATOMICAL LANDMARKS

The concept of CVS was introduced by Strasberg et al[1] in 1995 and was further emphasized as a method to avoid four common pitfalls encountered during LC: The infundibular view error trap, the fundus-down error trap, failure to recognize an aberrant right hepatic duct, and parallel union of the cystic duct (CD)[1,2]. To further enhance safety, “time-out” and “bailout techniques” were incorporated to define the concept of “culture of safety” in 2013[3]. Other anatomical landmarks that define safe limits of dissection include Rouviere’s sulcus (RS), the RS, segment 4, umbilical fissure (R4U) line, and the B-SAFE landmarks (bile duct, RS, hepatic artery, falciform ligament, and enteric structures)[4]. When visible, these landmarks help guide the limits of safe dissection zones. In the scenario of a simple, straight forward case with normal anatomy and minimal inflammation or adiposity, these principles can be applied easily. However, when anatomy is obscured or landmarks are not identifiable, the use of alternative strategies is advised.

TRADITIONAL INTRAOPERATIVE IMAGING AIDS: INTRAOPERATIVE CHOLANGIOGRAPHY AND INTRAOPERATIVE ULTRASOUND-ROLE IN DELINEATION OF BILIARY ANATOMY

Intraoperative cholangiography

Evidence from systematic reviews and meta-analyses suggest that intraoperative cholangiography (IOC) improves visualization of normal as well as aberrant biliary anatomy, is useful for the detection of common bile duct (CBD) stones and may decrease BDI rates. However, the routine use of IOC increases operative time and cost, and its superiority over selective IOC remains debatable. Moreover, it is invasive, exposes patients to radiation, has a learning curve, and requires initial dissection for CD cannulation, which can be challenging in a frozen Calot’s triangle[5-8].

Intraoperative ultrasound

Available evidence suggests that intraoperative ultrasound (IOU) is a safe and accurate modality for assessing biliary anatomy and detecting stones during LC. It offers some advantages over IOC, such as: Lower failure rates, shorter operative time, avoidance of radiation exposure, non-requirement for CD cannulation, and improved delineation of vascular anatomy. However, it has an appreciable learning curve, and current evidence does not conclusively demonstrate a reduction in BDI rates compared to IOC[7,9-12].

Anatomical mapping: Pre-dissection mapping of hepatobiliary anatomy is a key advantage of IOU and indocyanine green near infrared (ICG NIR) cholangiography, particularly in cases of frozen Calot’s triangle where unguided dissection for CD cannulation in IOC can risk ductal injury or avulsion[13,14].

Identification of aberrant anatomy: Both IOC and IOU demonstrate a significantly higher detection rate (10%) of anatomical variations compared to no imaging[15].

Safety and BDI prevention-routine vs selective use of IOC: Routine IOC use has been associated with a 29% reduction in BDI rates and a 62% reduction in mortality in large national registries, although, some studies show no significant difference between routine and selective use[15,16].

NEWER ENDOVISION-AIDED MULTISPECTRAL IMAGING-INDOCYANINEGREEN NEAR INFRA RED CHOLANGIOGRAPHY, YELLOW ENHANCEMENT, AND OTHER SPECTRAL MODES: ROLE IN DELINEATION OF HEPATOBILIARY ANATOMY

ICG-NIR

ICG-NIR cholangiography has emerged as an important adjunct for improving the safety of LC by enabling real-time visualization of biliary anatomy without radiation exposure. It reduces operative time, decreases conversion to open surgery, minimizes postoperative complications, and, in some studies, has been suggested to reduce BDI, primarily due to its superior ability to map biliary anatomy compared with white light imaging (WLI). CBD detection tunes to the rate of 78.6% vs 49.7% ICG-NIR vs WLI[17-21]. However, a recent meta-analysis of 1586 patients that included only RCTs found no significant difference in BDI rates[20]. Other meta-analyses with larger patient samples (3457 and 2490 patients, respectively) noted a trend toward lower BDI rates with ICG-NIR use, although the difference did not achieve statistical significance (ICG vs WLI: 0.018% vs 0.225% and 0.12% vs 1.31%, respectively, in the two studies)[18,22]. Another meta-analysis of 4436 patients reported a lower but statistically insignificant BDI rate with the use of ICG fluorescence cholangiography[21]. Thus, most studies suggest a trend toward lower BDI rates with ICG NIR than with WLI, which does not achieve statistical significance, presumably due to the low overall incidence of BDI.

Comparison with other modalities IOC/IOU vs ICG-NIR: For general biliary pathology (excluding stones), ICG-NIR is now considered non-inferior to IOC for critical extrahepatic biliary anatomy visualization. It is non-invasive, has a shorter learning curve, reduces procedure time, and offers higher surgeon satisfaction[14,22]. The SAGES 2025 guideline issued a conditional recommendation suggesting the “superiority of IOC over ICG” for reducing BDI rates, with low certainty of evidence[15]. Other RCTs with larger patient populations, including one with a three-arm comparison, reported equivalent results with the use of ICG-NIR and IOC for biliary anatomy visualization[14,23].

Role of ICG-NIR cholangiography in navigation of LC in easy and difficult cases: ICG-NIR cholangiography demonstrates high sensitivity for identification of key anatomical structures and shows clear superiority over WLI, particularly in cases involving inflammation or obesity. The use of NIR II window scores over NIR I in such cases[22,24].

In easy cases, ICG enables near-universal visualization of the CD-CBD junction. However, its true value, in terms of accuracy, is realized in difficult cases, where it outperforms WLI in anatomical precision, conversion rates, operative duration, and surgeon comfort. Even when the CD-CBD junction is not visualized, ICG NIR can help define safe zones of dissection by excluding the presence of the CBD in the operative field. Early identification of these safe zones can reduce operative time by an average of 20 minutes in inflamed or scarred Calot’s triangle[22]. A large retrospective study reported conversion rate of 1.2% with ICG NIR vs 3.3% without it and found it to be beneficial in fundus first technique[25]. In complex anatomical scenarios, surgeons also reported higher satisfaction and lower mental workload when using ICG NIR compared to IOC (4.1 vs 2.9 on a 5-point scale)[14]. CD identification rates range between 95%-100% vs 85%-93%, with the corresponding CBD identification rates being 96% vs 78%-86% in easy vs difficult cases respectively[14,25,26]. In contrast, ICG NIR offers no significant advantage over WLI in straightforward cholecystectomy, where conversion rates, anatomical identification, and CVS achievement are comparable[14,25,26].

Limitations of ICG-NIR cholangiography: Some of the noted limitations of ICG-NIR cholangiography are as follows: (1) Obesity, inflammation, and fibrosis may limit visibility; (2) Non-visualization of the gallbladder due to obstruction of the CD by a stone at the neck; (3) Dye spillage from gallbladder perforation may contaminate the operative field; and (4) Background fluorescence due to hepatic uptake may blur visualization.

ICG penetration in high body mass index and inflamed/fibrous tissue: The depth of penetration of ICG fluorescence is limited to approximately 1 cm and may be hindered by visceral fat or fibrotic tissue[14,26]. Although high body mass index (BMI) may impair early detection of biliary anatomy, identification rates of critical structures often become comparable to those in normal BMI patients after initial dissection. ICG-NIR use can reduce operative time and conversion rates in difficult cases. Identification of biliary anatomy is faster with ICG-NIR (approximately 110 seconds) compared to conventional IOC (approximately 305 seconds) in obese patients. Modified protocols for high-BMI patients-such as extended injection timing (3-6 hours preoperatively) and microdosing (0.05 mg/kg) can improve the signal-to-background ratio and prevent excessive hepatic fluorescence[27-29]. The evolving synergy between AI and ICG NIR may further enhance these outcomes.

Non enhancement of gallbladder in ICG-NIR due to blocked CD stone at the neck: A blocked CD due to fibrosis, scarring, or a large stone at the neck may lead to non-enhancement of the gallbladder. However, other critical anatomy, such as the CBD and CHD, may remain identifiable and can guide safe dissection boundaries. Alternatively, dye instillation directly into the gallbladder, preferably through a trans-hepatic route, can be used to circumvent this issue. The absence of hepatic uptake improves gallbladder visualization by decreasing the signal-to-noise ratio; however, some caveats of the procedure include the light bulb effect due to high ICG concentration in the gallbladder, accidental spillage of dye into the operative field impeding visualization, and inability or incorrect identification of the CBD due to a blocked CD or pseudo-CVS caused by a tented CBD[30,31].

Background noise due to hepatic uptake: Emerging evidence suggests that adjustments in dosage and timing of ICG administration, use of NIR window II, and deployment multicolor fluorescence with toggling may improve the clarity of biliary anatomy visualization by decreasing the signal-to-noise ratio[24,27-29,32].

Yellow enhancement and other spectral modes

Newer endovision systems incorporate additional spectral modes that may aid LC: (1) Olympus yellow enhancement (YE): Accentuates fat–duct contrast, may improve visualization of cystic artery; (2) Stryker (tone/color spectral imaging): Enhances “two-structure” visualization by improving vascular ductal contrast; and (3) Karl Storz (Clara/Chroma): Improves visualization of the cystic plate by illuminating deep shadows in the liver bed.

Software-based filters such as YE can increase fat differentiation in endovision imaging systems, allowing tissue-specific mapping. YE has been suggested to improve tissue visualization in obese patients and in cases of chronic inflammation[33]. In difficult gallbladders with fat-laden pedicles or fibrosis in the region of the hepatocystic triangle (HCT), use of YE has been proposed to improve visualization of avascular planes of dissection and decrease operative time by 17%-19%[34].

Limited studies have reported on role of YE in LC cholecystectomy. A study comparing YE with WLI using a 5-point Likert scale in cholecystectomy and hernia surgery noted improved clarity in the identification of vascular structures (4.8 vs 3.2), fat differentiation (4.7 vs 2.9), avascular plane detection (4.9 vs 3.1) and surgeon satisfaction[34]. Toggling between different modes, such as YE and ICG-NIR, may prove helpful in difficult cases[35]. However, data remain sparse but evolving, and the final impact of these spectral modes on outcomes such as BDI, bleeding, hospital stay, and complications need to be evaluated. Figure 1 depicts operative photographs of LC with dissected HCT and achievement of CVS, as observed in WLI, YE, and ICG-NIR modes, respectively.

Figure 1 Three snapshots (deidentified operative photograph) of laparoscopic cholecystectomy with dissected hepatocystic triangle and critical view of safety achievement as observed. A: White light imaging; B: Yellow enhancement (YE); C: Indocyanine green near infrared fluoroscopy modes (ICG NIR). Note improved fat differentiation with YE and precise common bile duct visualization (black arrow) in ICG NIR modes.

ROLE OF AI IN INCREASING PRECISION AND SAFETY OF LC

AI models used in LC primarily rely on the following technologies: Computer vision, machine learning/deep learning, neural networks, and video analytics. Table 1 shows the technical process and typical use of common AI architecture used in LC.

Table 1 Technical process and typical use of common artificial intelligence architectures used in laparoscopic cholecystectomy.

AI architecture	Technical process	Primary logic	Thinking style	Typical use
CNN	CV and DL	Spatial. Image analysis	What is the object?	Sees pixels and identify shapes-organ/tool detection/segmentation
Transformer	DL and video analytics	Temporal dependency. Sequence. Video analysis	What happens next?	Looks at the sequence of events over time-video analysis, workflow assessment and phase recognition
GNN	DL and Topological AI	Relational/topological	How are these connected?	Goes beyond seeing pixels to structure-CVS and anatomy mapping
Foundational	Generalist AI/MLLMs	Knowledge	I have seen this before	Generalizable feature extraction and cross-procedural understanding

CVS: Critical view of safety; CNN: Convolutional neural network; GNN: Graph neural network; CV: Computer vision; DL: Deep learning; MLLMs: Multimodal large language models; AI: Artificial intelligence.

AI is being applied in LC across several domains: (1) Operative complexity/difficulty prediction and grading systems- preoperative difficulty prediction and intraoperative difficulty grading; (2) Surgical workflow assessment and phase recognition; (3) Anatomic identification and segmentation of structures such as HCT, CBD, gallbladder etc.; (4) Delineation of Go and No-Go zones; (5) Assessment of CVS achievement; (6) Safety assessment and- instrument tracking; (7) Context-aware surgery; and (8) Coaching and training.

AI-enabled prediction and scoring systems for difficult cholecystectomy: Accurate difficulty prediction of operative difficulty using preoperative and intraoperative parameters is essential for improving safety in LC. Such predictions can guide the use of adjunctive imaging modalities, workflow optimization, and timely adoption of bailout strategies. Most traditional severity grading systems lack accuracy and external validation, necessitating the development of AI-based algorithms. Current AI models demonstrate modest to high accuracy (72%-98%) in predicting difficult LC and assessing operative severity, including Parkland grading[36-41].

Surgical phase recognition and workflow assessment: AI models have demonstrated substantial accuracy in identifying phases of LC, viz., trocar placement, Calot’s dissection, clipping, etc. Reported accuracy ranges from 80% to 90%, depending on case complexity, with some studies showing near-complete agreement with expert assessment in real time. Automated workflow and phase recognition can standardize surgical processes, identify deviations from optimal sequences, and enable context-aware assistance by generating alerts during high-risk phases[42,43].

AI-based recognition of surgical anatomy and CVS

Anatomical recognition and CVS assessment: Various segmentation techniques are being explored for anatomical identification: Bounding boxes: Used for object detection, broadly identifying location of major organs or areas of interest (Figure 2A). Semantic segmentation: Pixel-wise classification to define anatomical boundaries (Figure 2B). Semantic and instance segmentation: Used for detailed assessment, including CVS evaluation and AI workflow pipeline (Figure 3). Go/No-Go zone mapping uses heatmaps to provide semi-transparent overlays that guides safe dissection limits (Figure 4).

Figure 2 Schematic representation of artificial intelligence-driven anatomic delineation of hepatocystic triangle using bounding box and semi-transparent overlay mask. A: Initial bounding box method (red rectangle) for broad target area identification and a confidence call out; B: Semantic segmentation further precisely defining the exact region of hepatocystic triangle (HCT) by using a semi-transparent optical density mask overlay.

Figure 3 Schematic representation of artificial intelligence processes for anatomical mapping of hepatocystic triangle and critical view of safety assessment using instance segmentation and overview of artificial intelligence workflow pipeline. A: Integrated artificial intelligence (AI)-enhanced anatomical mapping with transparent mask overlay. This schematic demonstrates the application of landmark-based anatomical mapping used for anatomical segmentation. Numbered identifiers represent key anatomical landmarks that can be used (e.g., 1: Cystic duct-gallbladder junction; 2: Cystic duct-common bile duct junction; 3: Cystic artery; 4: Hepatocystic triangle; 5: Cystic plate; 6: Liver). The figure proposes the concept of using a transparent mask with borders for delineating the hepatocystic triangle (red boundary line). This approach minimizes the obscuration of underlying anatomical textures compared to traditional opaque overlays. The textbox shows the confidence callout of the model in identifying hepatocystic triangle. Numerical labels are illustrative and do not correspond to a specific validated system; B: Instance segmentation for anatomical decomposition and critical view of safety assessment. This schematic illustrates the application of instance segmentation to identify and differentiate individual anatomical structures within the hepatocystic triangle, shown in a spatially decomposed arrangement. Unlike semantic segmentation, which assigns a single class label to all pixels of a given structure type, instance segmentation enumerates discrete anatomical objects individually. This enables automated verification of the critical view of safety, providing an algorithmic baseline to assess the “two-structure rule” (Cystic duct and Cystic artery entering the gallbladder) prior to clip application and ductal division. Numerical labels are illustrative. Clinical application requires surgeon confirmation of model outputs; C: AI pipeline for segmentation and anatomical analysis. Laparoscopic video/still images of the hepatobiliary anatomy are passed to an AI based segmentation model, which extracts hierarchical anatomical features via a convolutional neural network backbone (represented by the double helix motif). The model generates a per-pixel classification map in which each pixel is assigned a discrete tissue or landmark label (represented by the color-coded checkered segmentation mask). Downstream outputs include instance-level identification of individual structures and overlay on the operative image, followed by 3D volumetric reconstruction of the surgical field to support spatial orientation. Icons are schematic representations of computational processes. AI: Artificial intelligence.

Figure 4  Schematic representation depicting mapping of Safe (Go) and unsafe (No-Go) zones in hepatocystic triangle (please note this is an illustration to show how heat maps are being used to delineate these zones and not an accurately scaled image).

Several open-source datasets (e.g., Endoscapes, CholecT50, CholecT80) have been developed to train AI models. While AI accuracy is improving, it still falls short of expert-level annotation and agreement. Additionally, AI can be trained to identify if CVS has been achieved as per Strasberg’s criteria. The accuracy of detecting anatomical landmarks in LC using a segmentation technique is variable dependent largely on AI model used, being lower in still images and smaller structures like CD/cystic artery vs gallbladder and improves with use of video analytics. In earlier studies accuracy level of 71.9% and a precision of 71.4% was noted[43]. However, recent studies noted improvement in CVS achievement with reported correct detection in 92% cases[44].

AI recognition of other anatomical landmarks RS/R4U line: Beyond the HCT, AI can also identify landmarks such as RS and the R4U line, further aiding safe dissection[45].

The Go vs the No-Go zones: Although achieving CVS is the gold standard in LC, it may be difficult or unsafe in scenarios such as a frozen Calot’s triangle. In such cases, adherence to safe dissection within predefined “Go zones” becomes critical.

AI in the detection of safe zones of dissection-Go/No-Go zones: AI-based systems are being developed for real-time identification of safe and unsafe dissection zones: The GoNo-GoNet algorithm provides semi-transparent overlays and heat maps over the HCT during critical phases of LC to guide dissection and has demonstrated a mean accuracy of 92% and specificity of 97% for identifying safe zones[46,47]. In a video analysis of BDI cases, it was observed that 36% of surgical interactions occurred in No-Go zones, 18% in Go zones, and 45% outside defined zones[48].

Other newer models under evaluation: Cross AI models detecting work phase as well as anatomical landmarks are evolving[42]. A recent study describes SurGo-R1 model which incorporates reinforcement learning from human feedback to reduce false positives in high-fat environments by first identifying the surgical phase before recommending dissection pathways. However, the report is currently in preprint stage and any conclusion in this regard is preliminary[49].

Synergy of AI and ICG AI-enabled hepatobiliary anatomic delineation in ICG modes: Most AI-based segmentation models have traditionally been developed using WLI; however, emerging data suggests improved performance with ICG-enhanced imaging. ICG-integrated AI demonstrates higher precision in anatomical delineation, particularly for identifying the CBD in high-BMI patients during the pre-dissection phase. Combining ICG with deep learning models such as you only look once (YOLO) significantly enhances detection accuracy as AI can identify subtle fluorescence patterns through 5-10 mm of adipose tissue that may not be visible to the human eye[50]. Reported accuracies include 94.39% for CBD detection and 84.97% for CD identification using YOLO v5 and v7 in ICG-based models[51]. A high no go specificity in the range of 95%-97% has also been noted along with real time latency like that of non-AI mode (< 30 ms)[51]. Table 2 shows the detection rate of the CD and CBD across Nassar grades in two independent non-comparative studies: Low-dose ICG-NIR in one study[52] vs AI-integrated ICG-NIR mode in the other study[51].

Table 2 Detection rate of cystic duct and common bile duct across Nassar grades in low dose indocyanine green near infrared vs artificial intelligence integrated indocyanine green near infrared modes (the two studies are independent non comparative studies).

Anatomical structure	ICG NIR FC[52]		AI enabled ICG NIR FC[51]
	Pre-dissection detection percentage	Post-dissection detection percentage	Video enabled detection percentage (sensitivity)
Nassar grade 1	CBD 39, CD 0	CBD 100, CD 70	CBD 100, CD 94.12
Nassar grade 2			CBD 98.61, CD 87.50
Nassar grade 3	CBD 0, CD 0	CBD 100, CD 73	CBD 100, CD 92.50
Nassar grade 4			CBD 98.25, CD 60.61

CBD: Common bile duct; CD: Cystic duct; ICG NIR FC: Indocyanine green near infrared fluorescence cholangiography; AI: Artificial intelligence.

AI in coaching and training: AI models can be leveraged to assist with the coaching and training of surgeons to perform LC. Limited evidence suggests a positive impact on surgical performance. Gamified learning platforms, such as LapBot, may enhance engagement and promote experiential learning in surgical training[53-55].

AI models can further be classified as per maturity levels into the following tiers: Tier 1, relatively mature tasks: Phase recognition, instrument detection, gallbladder and hepatocystic anatomy segmentation. Tier 2, tasks under validation: CVS documentation, Go/No-Go zone mapping, operative difficulty grading. Tier 3, early proof-of-concept: ICG-enhanced AI bile duct detection, context-aware decision support, automated coaching. Tier 4, future directions: GNN-based relational anatomy mapping, foundation-model generalization, autonomous robotic platforms. Table 3 summarizes utility and limitations of various imaging aids, endo vision systems and AI in detection of biliary anatomy and pathology in LC.

Table 3 Utility and limitations of various imaging aids, endo vision systems and artificial intelligence in detection of biliary anatomy and pathology in laparoscopic cholecystectomy.

Mode	CBD detection rates	Benefits	Limitations
WLI	49.7%[17]	No additional training or equipment is needed	Low sensitivity in difficult grades
IOC	87% to 95%[10]	Good for biliary anatomy particularly aberrant ducts. Good visualization of entire biliary tree. Detection of simultaneous CBD stones. Simultaneous intervention possible for stones	Learning curve, need for CD cannulation, radiation exposure, visualization is not real time, not for vascular anatomy delineation, false positives, routine use advocated can increase time and cost, not repeatable
IOU	92%-100%[13]	Good for biliary anatomy particularly aberrant ducts. Detection of simultaneous CBD stones. Good for vascular anatomy delineation. No CD cannulation needed. No risk of radiation or dye allergies. Can be repeated easily	Steep learning curve. Operator dependent. Difficulty visualizing distal CBD. Inflammatory interference. Limited “Map” of injury
ICG NIR	96% vs 78%-86% in easy vs difficult cases[14,25,26]	Short learning curve, no radiation exposure, real time visualization, no CD cannulation, intraoperative navigation of safe dissection zones	Dye spillage from bile leak/GB perforation can impede visualization. Non-visualization-blocked CD/stones. Hepatic fluorescence interference. Penetration limited in fat/inflammation. Not useful for CBD stones. Small chance of dye allergy
YE and other spectral modes	No concrete data available presumed to be same as WLI	Improved vascular anatomy detection particularly cystic artery. Detection of cystic plate. Rouviere’s sulcus and other surface topography	Role still evolving, limited evidence only few case reports
AI in WLI	71% anatomy detection 92% CVS detection 92% accuracy for safe zones[44,46,47]	Moderate sensitivity for anatomy detection. High sensitivity for CVS detection	Needs validation in further studies. Requires AI enabled endo vision systems for real time mapping
AI and ICG integration	94.39% with YOLO. 82%-100% depending on difficulty grades[51]	High sensitivity in difficult grades. May be helpful for real time definition of safe dissection zones	Needs validation in further studies. Requires AI enabled endo vision systems for real time mapping

CBD: Common bile duct; CD: Cystic duct; GB: Gallbladder; CVS: Critical view of safety; WLI: White light imaging; IOC: Intraoperative cholangiography; IOU: Intraoperative ultrasound; ICG NIR: Indocyanine green near infrared fluorescence; YE: Yellow enhancement; AI: Artificial intelligence; YOLO: You only look once.

LIMITATIONS AND FUTURE DIRECTIONS

Despite the promise of AI and multispectral imaging, several challenges remain to be addressed, including: (1) Learning curve associated with AI-integrated systems; (2) Performance in cases with dense inflammation or significant obesity; (3) Generalizability and broader applicability; (4) Cost-effectiveness; and (5) Safety and regulatory issues. Future developments in AI applications may include but not limited to: (1) Integration with advanced multispectral imaging; (2) Enhanced precision in real-time surgical navigation; (3) Context-aware surgical assistance; (4) Standardized AI-driven coaching systems; and (5) Development of autonomous robotic platforms.

CONCLUSION

Surgical precision refers to the ability to perform procedures with high accuracy and minimal deviation from the intended target area. Accurate delineation of anatomic structures remains the first step in ensuring precision. Integration of traditional systems such as IOC and IOU with advanced endovision modes such as ICG-NIR, YE and AI can provide improved tissue differentiation and anatomical demarcation compared with WLI and has been suggested to improve precision. Until recently, AI models were being evaluated largely in non-real-time scenarios. Exploration of nascent AI architectures, such as CNN and transformers, has the potential to assist in real-time delineation of critical anatomy, mapping of safe dissection zones across different dissection phases, and generation of safety alerts in situations involving dangerous tissue interactions during surgery. Use of these models and systems should be encouraged within ethical and regulatory boundaries to validate their role in improving the precision and safety of LC. Despite these advances, bailout strategies such as the fundus-first technique and subtotal cholecystectomy remain cornerstones in the management of the difficult cholecystectomy.