Artificial Intelligence in Surgical Gastroenterology: From Predictive Models to Intraoperative Guidance

Abstract

Background:

Artificial intelligence (AI) is transforming surgical gastroenterology by enabling data-driven precision across the preoperative, intraoperative and postoperative continuum. This review aims to synthesise recent advances and evaluate the clinical applicability of AI-powered platforms in gastrointestinal (GI) surgery.

Objectives:

To provide a comprehensive overview of AI integration in surgical gastroenterology between 2015 and 2025, focusing on predictive analytics, intraoperative guidance, postoperative surveillance, implementation challenges and the Indian context.

Methods:

A narrative review was conducted using the PubMed, Scopus and Web of Science databases. Studies published between January 2015 and May 2025 were included. AI domains covered include machine learning (ML), deep learning (DL), natural language processing (NLP), computer vision and augmented reality (AR). Only peer-reviewed, English-language articles relevant to surgical GI practice were selected.

Results:

AI models have demonstrated superior accuracy over traditional scoring systems in predicting anastomotic leaks (ALs), readmissions and postoperative liver failure. Intraoperatively, real-time computer vision platforms enable anatomical recognition, tool tracking and perfusion analysis. Postoperative monitoring systems, such as FluidAI and MySurgeryRisk, provide early complication alerts through multimodal data integration. Despite the promise, challenges persist, including data heterogeneity, limited external validation, algorithmic bias and regulatory ambiguity. A tailored roadmap for Indian healthcare outlines priorities in dataset development, ethics frameworks and capacity building.

Conclusion:

AI is poised to augment decision-making and improve outcomes in surgical gastroenterology. Its successful adoption depends on equitable data access, transparent algorithms and surgeon-led innovation, particularly in resource-diverse settings such as India.

Keywords

Artificial intelligence gastrointestinal surgery machine learning intraoperative guidance postoperative monitoring India surgical outcomes

Introduction

Over the last decade, artificial intelligence (AI) has transformed from an experimental novelty into a disruptive force in surgical gastroenterology. Complex gastrointestinal (GI) surgeries, ranging from colorectal resections to hepatopancreatic procedures, demand intricate decision-making across the preoperative, intraoperative and postoperative continuum. Traditional clinical tools, including risk calculators and scoring systems, often rely on linear statistical models and a limited array of variables, rendering them inadequate in the face of nonlinear patient data and dynamic intraoperative environments.^[1,2]

AI, encompassing machine learning (ML), deep learning (DL), natural language processing (NLP) and computer vision, offers an unprecedented ability to process high-dimensional data streams from clinical records, imaging, operative videos and sensor-based feeds.^[3,4] By learning complex, multivariate associations, AI-driven algorithms can predict surgical outcomes, support intraoperative navigation, personalise postoperative surveillance and even assist in surgical training.^[5,6] For instance, AI models have demonstrated superior accuracy in forecasting anastomotic leaks (ALs) and postoperative complications compared to conventional methods, while augmented reality (AR) and real-time video analytics now facilitate anatomical guidance and instrument tracking in real-time.^[7,8]

Despite its promise, the integration of AI into surgical gastroenterology faces several challenges, including data heterogeneity, lack of external validation, ethical considerations and regulatory ambiguity.^[9] Additionally, questions remain regarding the interpretability and generalisability of AI-driven skill assessment models in real-world surgical education settings.^[10] Nonetheless, the rapidly expanding body of high-quality evidence signals a paradigm shift toward data-informed, algorithm-enhanced GI surgery.

A timeline of key milestones highlighting the transformation of AI in surgical gastroenterology over the last decade is summarised in Table 1.

Table 1.

Evolution of AI in Surgical Gastroenterology (2015–2025)

Timeline of key technological transitions, clinical milestones and paradigm shifts
Year	Milestone / Innovation	AI Domain	Clinical Relevance
2015	Early risk prediction using logistic regression and ML hybrids (e.g., RF and SVM)	Predictive analytics	Initiated transition from static risk calculators to dynamic models
2017	First prospective studies integrating operative video and time-series data	Deep learning (DL), recurrent neural networks (RNN)	Enabled intraoperative event forecasting (e.g., AL, bleeding)
2018	Launch of MySurgeryRisk EHR-based complication predictor	Structured data AI	Demonstrated superior accuracy to clinician judgement for AKI/sepsis
2019	ICG perfusion quantification using feed-forward NNs in colorectal surgery	Computer vision	Improved leak risk stratification during anastomosis creation
2020	Integration of convolutional neural networks (CNNs) with laparoscopic video	Vision-based AI	Enabled real-time tool tracking and phase recognition in GI procedures
2021	Hepataug AR system validated for liver resection navigation	Augmented reality (AR)	Reduced blood loss and improved intraoperative spatial orientation
2022	LightGBM and ensemble models surpass MELD in Post-hepatectomy liver failure (PHLF) prediction	Ensemble ML	Outperformed traditional hepatic scoring systems in accuracy
2023	AiLES system matches novice performance in detecting gastric metastases	Real-time image analytics	Advanced role of AI in pre-resection staging and metastasis mapping
2024	Quantum neural networks piloted for complication prediction	Hardware-accelerated AI	Explored scalable AI inference models for bedside use
2025	Federated learning, explainable AI frameworks (SHAP) begin real-world deployment	Ethics-integrated AI	Improved model generalisability, accountability and fairness

AI Domain: The specific subfield of AI applied, including ML, DL, CNN, recurrent neural networks (RNN) and AR.

Predictive analytics: Algorithms designed to forecast outcomes such as complications, mortality or readmission using structured clinical data (laboratories, vitals, demographics).

Computer vision: AI methods that analyse surgical images or video for anatomical recognition, tool tracking or perfusion assessment.

Augmented reality (AR): Technology that overlays digital content (e.g., 3D tumour models) onto the surgical field to assist with navigation.

MySurgeryRisk: A validated ML platform trained on EHR data to predict major postoperative complications in real-time.

ICG: Indocyanine Green; a fluorescent dye used intraoperatively to assess perfusion, now enhanced by AI classification.

LightGBM: A gradient boosting decision tree algorithm optimised for speed and performance, often used in clinical risk modelling.

AiLES: Artificial Intelligence Laparoscopic Exploration System, used for intraoperative detection of peritoneal metastases.

Quantum neural networks: A next-generation AI framework utilising quantum computing principles to enhance processing efficiency.

SHAP (SHapley Additive exPlanations): An explainability framework used to understand the contribution of individual features to an AI model’s prediction.

This narrative review aims to synthesise key developments in AI applications for surgical gastroenterology between 2015 and 2025.

We examine five pivotal domains: (a) Predictive analytics for perioperative decision-making, (b) intraoperative AI for navigation and decision support, (c) postoperative complication monitoring, (d) AI-enhanced surgical education and (e) translational challenges and future directions. By critically evaluating both the technological evolution and clinical validation of these innovations, we offer a consolidated reference point for clinicians, researchers and policymakers navigating the AI-enabled surgical frontier.

Methods

This narrative review was conducted in accordance with the scale for the assessment of narrative review articles (SANRA) guidelines. A literature search was performed using PubMed, Scopus and Web of Science databases to identify relevant publications from January 2015 to May 2025. Search terms included combinations of ‘AI,’ ‘ML,’ ‘DL,’ ‘computer vision,’ ‘surgical gastroenterology,’ ‘predictive analytics’ and ‘AR’. Only peer-reviewed, English-language studies were considered. Reference screening and inclusion prioritisation were guided by clinical relevance, technical rigour and surgical applicability.

Predictive Analytics in Surgical Gastroenterology

The preoperative and perioperative phases in GI surgery are fraught with uncertainty, ranging from patient selection to predicting complications. AI, particularly ML and DL, has enabled a shift from static risk stratification to dynamic, individualised prognostication. AI-driven models now utilise multidimensional datasets including laboratory trends, operative timings, radiomics and sensor data to estimate postoperative outcomes with increasing precision.

In colorectal surgery, the prediction of AL has been a cornerstone application. A multicentre analysis of AI models, including random forests and neural networks, reported area under the receiver operating characteristic curve (AUROC) up to 0.83 across institutional datasets, though external validation often revealed performance drops, highlighting the challenge of model generalisability.^[11] AI models integrating auto-selection of circular stapler types and perfusion maps using indocyanine green (ICG) with feed-forward neural networks improved AL prediction accuracy to an AUROC of 0.774 by incorporating real-time operative features.^[12]

Beyond binary outcome forecasting, AI is being used for unsupervised clustering of postoperative complications. A meta-analysis of 38 studies between 2015 and 2022 found that ML methods improved mean area under the curve (AUC) by 0.07 compared to traditional regression-based models for endpoints such as 30-day readmission and postoperative ileus.^[13] For example, a support vector machine (SVM) model using postoperative day three C-reactive protein (CRP), length of stay and creatinine levels achieved an AUROC of 0.784 for readmission, outperforming conventional multilayer perceptrons.^[14]

In gastric cancer surgery, hybrid convolutional neural network (CNN)-Transformer models have demonstrated 90% accuracy in recognising surgical phases, enabling predictive guidance for intraoperative events such as vessel control or stapler deployment.^[15] Similarly, in pancreatic surgery, gradient boosting methods have surpassed logistic regression in predicting futile pancreaticoduodenectomy, defined as death within 12 months, achieving an AUROC of 0.689 versus 0.679.^[16] Clinically relevant postoperative pancreatic fistulas have also been predicted using ensemble radiomics and DL models with an AUROC of 0.759.^[17]

Hepatic procedures have witnessed a parallel evolution. A LightGBM-based risk model developed on 935 patients achieved an AUROC of 0.812 (95% CI, 0.791–0.853) in predicting post-hepatectomy liver failure, outperforming model for end-stage liver disease (MELD) and albumin-bilirubin index (ALBI) scores.^[18] Meanwhile, DL models analysing surveillance imaging and laboratory data have achieved concordance indices as high as 0.83 for five-year hepatocellular carcinoma (HCC) recurrence, well above traditional Cox models.^[19]

In bariatric surgery, SVMs leveraging perioperative laboratory parameters have shown promise in forecasting 30-day readmission, reaching an AUROC of 0.784 in large-scale US datasets.^[20]

A comparative overview of organ-specific AI models, including outcomes, algorithmic performance and validation status, is provided in Table 2.

Table 2.

Representative predictive AI models in GI surgery (2015–2025)

Organ-wise summary of AI applications for outcome prediction with technical and clinical performance parameters
Organ	Predicted Outcome	Algorithm / Model	AUROC / Accuracy	External Validation	Key Predictors
Colon and Rectum	Anastomotic leak (AL)	Random forest (Swiss cohort)	0.783 → 0.60 (external)	✅	Malnutrition, immunosuppression and male sex
	30-day readmission	Support vector machine (SVM)	0.784	❌	CRP (POD3), LOS and creatinine
Stomach	Surgical phase recognition	CNN and transformer	90% (accuracy)	N/A	Temporal-spatial video features
	AL post-gastrectomy	RF WebApp (Korean study)	0.77	❌	ASA grade and operative time
Pancreas	Futile PD (12-mo mortality)	Gradient boosting	0.689	❌	Age, BMI, vascular resection and bilirubin
	Post-op pancreatic fistula	Ensemble DL and radiomics	0.759	❌	Duct size, gland texture and blood loss
Liver	PHLF	LightGBM	0.822	✅	INR, ICG-R15 and ALBI score
Bariatric	30-day readmission	SVM (US cohort)	0.784	❌	Haemoglobin, creatinine and operative time
HCC (Surveillance)	5-year recurrence	Deep neural net	C-index 0.83	✅	AFP, tumour size and vascular invasion

Note: AUROC: Area under the receiver operating characteristic curve. Values closer to 1.0 indicate better model performance; C-index: Concordance index used in survival models (similar to AUROC); SVM: Support vector machine, a supervised ML classifier; CNN: Convolutional neural network, commonly used for image/video data; LightGBM: Gradient boosting decision tree framework, high performance for tabular data; DL: Deep learning, encompassing multilayered neural networks; ICG-R15: Indocyanine Green retention at 15 minutes; liver function test; AFP: Alpha-fetoprotein; HCC surveillance biomarker; ALBI Score: Albumin-bilirubin index; liver function risk stratifier; POD3: Postoperative Day 3; LOS: Length of stay; ASA: American society of anesthesiologists; BMI: Body mass index.

These advances underscore AI’s transition from theoretical modelling to clinical toolkits. Yet, a common limitation remains the lack of multicentre validation, which often leads to reduced accuracy outside the development cohort. Bridging this gap will be key to translating predictive analytics into daily surgical decision-making.

Intraoperative AI and Computer Vision in GI Surgery

The intraoperative environment presents a high-stakes, time-critical landscape where real-time decision-making can significantly alter outcomes. AI, particularly computer vision and AR, is increasingly being deployed in GI surgery to enhance anatomical recognition, optimise tool handling and support surgical judgement. These systems leverage live video, sensor fusion and trained neural networks to provide immediate actionable insights within the sterile field.

In hepatic surgery, AR platforms such as Hepataug overlay 3D tumour reconstructions onto the surgeon’s laparoscopic view. In one validation, physics-based deformation mapping using red-green-blue plus depth camera input (RGB-D) sensors achieved a mean projection error of just 9.2 mm, compared to 29.4 mm with traditional laparoscope-axis alignment.^[6] The clinical relevance of these overlays is underscored by reductions in transfusion rates and operative time.^[21]

AR’s utility has also been explored in pancreaticoduodenectomy, where systems such as augmented reality–assisted autonomic nerve segmentation (AR-ANS) enabled manual 3D holographic reconstructions to assist with margin identification and vascular orientation. A pilot study found AR integration correlated with a 35% reduction in intraoperative blood loss, although quantitative error metrics were not uniformly reported.^[22]

In colorectal surgery, perfusion assessment is an established predictor of anastomotic integrity. A quantitative AI model analysing ICG fluorescence angiography enabled accurate perfusion classification and leak prediction, marking a significant advance in intraoperative assessment.^[23]

Autonomic nerve preservation during pelvic dissection is another domain where AI has proven advantageous. DL-based segmentation of nerve plexuses during laparoscopic colorectal surgery achieved a Dice coefficient of 0.56, with earlier recognition than the operating surgeon in 50% of reviewed cases.^[24]

Tool tracking and segmentation have also advanced, with AI platforms capable of recognising multiple instruments and actions in real-time surgical video. A 2,024 systematic review summarised these efforts, emphasising improved detection accuracy, real-time processing and validation across real-world datasets.^[25]

Fluorescence-based perfusion evaluation has now reached randomised trial validation. In the Indocyanine Green–Enhanced Colorectal Angiography Trial (ICG-COLORAL) trial, AI-enhanced fluorescence imaging significantly reduced the rate of ALs in colorectal surgery by improving intraoperative tissue viability assessment.^[26]

Meanwhile, platforms such as AI laparoscopic exploration system (AiLES) have shown promise in metastasis detection during gastric staging laparoscopy, matching novice performance while operating at 11 frames per second.^[27]

In robotic surgery, total tool motion extracted from console data correlates strongly with surgical efficiency, with Pearson coefficients of −0.72 in simulation environments.^[28] CNNs trained on kinematic profiles distinguish expert from novice performance with >97% accuracy, offering potential for standardised, automated skill assessment.

Final, reinforcement learning (RL) is emerging for semi-autonomous tasks. Systems optimised through RL have demonstrated reduced force application and enhanced accuracy in robotic suturing, hinting at the future of intelligent surgical automation.^[29]

The major intraoperative AI systems in current surgical practice, along with their modalities, functionality and technical performance, are outlined in Table 3.

Table 3.

Intraoperative AI systems in gastrointestinal surgery

AI-based modalities enhancing real-time surgical performance and precision
Modality / System	Procedure / Target	AI Functionality	Performance Metrics	Clinical Impact
Hepataug (AR)	Laparoscopic liver resection	3D tumour overlay via RGB-D mapping	Mean error: 9.2 mm	Reduced transfusion and better spatial orientation^[6,21]
AR-ANS (Manual AR)	Pancreaticoduodenectomy	3D hologram guidance (manual overlay)	Qualitative	↓ Blood loss by 35%, improved vascular visualisation^[22]
ICG-NN System	Colorectal surgery	Perfusion adequacy classification (ICG fluorescence)	Accuracy: 99.9%	Leak risk reduction and better anastomotic viability^[23]
Nerve Segmentation DL	Laparoscopic colorectal	Autonomic nerve identification	Dice coefficient: 0.56	Earlier recognition versus the surgeon in 50% cases^[24]
AiLES	Gastric staging laparoscopy	Metastasis detection (image-based)	Dice: 0.76; 11 fps	Comparable to novice-level detection; faster diagnostics^[27]
Tool Recognition (DL-based)	Robotic/laparoscopic GI surgeries	Instrument tip detection and segmentation	Multi-centre accuracy >95%	Skill scoring, workflow optimisation^[25,37]
GEARS-Correlated Motion Metrics	Robotic simulation	Tool motion tracking and efficiency scoring	r = –0.72 (inverse correlation)	AI-assisted skill benchmarking^[28]
RL-based Suturing Systems	Bench simulation	Reinforcement learning-guided control	↓ Error and ↓ force application	Toward semi-autonomous task execution^[29]

Note: AR: Augmented Reality; RGB-D: Red-Green-Blue-Depth camera input; ICG: Indocyanine Green, used for real-time perfusion visualisation; Dice coefficient: Measures spatial overlap between AI prediction and ground truth (1.0 = perfect match); fps: Frames per second, indicates real-time processing capability; r (correlation): Strength of the relationship between tool motion and surgical efficiency; RL: Reinforcement Learning, an AI method that improves via trial-feedback loops.

These innovations collectively suggest that intraoperative AI is rapidly evolving from an assistive modality to a context-aware co-surgeon, redefining precision in GI s urgery.

Postoperative Monitoring and ‘Surgomics’

The postoperative phase remains one of the most unpredictable and resource-intensive components of GI surgery. Traditional surveillance protocols often rely on clinical suspicion, delayed imaging or reactive laboratory markers. AI introduces a proactive paradigm, where continuous data streams from surgical drains, vital signs and electronic health records (EHR) are leveraged to forecast complications such as ALs, surgical site infections and multi-organ dysfunction.

Among the most innovative tools is the FluidAI Stream™ platform, which integrates biochemical analysis of drain effluent with ML algorithms to flag ALs up to six days before clinical signs emerge. This early warning system has enabled both pre-emptive interventions and expedited discharge in low-risk patients.^[30]

Complementing biochemical analysis is an EHR-integrated AI platform developed by Xu et al. which has shown AUROCs exceeding 0.85 for predicting sepsis, acute kidney injury and respiratory failure in postoperative patients.^[31] By continuously updating predictions based on real-time data, such systems embody the emerging field of ‘surgomics’, the extraction of high-dimensional phenotypes from perioperative environments.

Quantum neural networks are being trialled for postoperative complication forecasting, offering comparable accuracy to classical models but with improved computational efficiency and adaptability.^[32] Though still in early-stage validation, they may enable bedside-level AI deployment using compact, edge-optimised hardware.

Multisensor integration is also advancing. Systems combining drain chemistry, thermal imaging and wearable biosensors have been shown to identify early signs of sepsis or wound dehiscence within 48 hours of surgery, well before standard monitoring would trigger concern.^[33] These platforms rely on feature fusion, where structured (laboratories, vitals) and unstructured data (free-text progress notes) are combined through DL to improve predictive power.^[33,35]

A prospective study by Nema and Vachhani demonstrated that integrating EHR data streams with NLP of operative notes improved the accuracy of predicting readmissions and delayed complications in pancreatic surgery.^[34] In parallel, AI-powered video analytics have also been used to flag intraoperative events predictive of postoperative complications, offering another layer of surveillance based on real-time behaviour analysis.^[35]

Importantly, AI not only detects deviations but helps define ‘normal’ recovery. A recent study mapped ideal postoperative trajectories across various GI procedures, creating dynamic baselines against which individual recovery curves could be compared. When embedded in clinical workflows, these baselines provided early alerts for outliers, prompting targeted follow-up.^[36] Recent advancements in real-time laparoscopic video analysis further emphasise the role of DL in instrument tracking, activity recognition and anomaly prediction.^[37,38]

Despite the promise, implementation remains limited by data silos, inconsistent documentation practices and privacy concerns. However, with the rise of federated learning models and real-time data interoperability standards, these barriers are gradually being eroded. Meanwhile, deep RL continues to evolve as a method for optimising postoperative pathways and autonomous clinical decision support systems.^[39]

Collectively, AI-enabled postoperative monitoring transforms passive recovery into active surveillance, enabling earlier rescue, personalised follow-up and data-informed discharge planning.

A summary of AI platforms enhancing postoperative surveillance, their data sources, clinical targets and readiness levels is presented in Table 4.

Table 4.

AI-based platforms for postoperative monitoring in GI surgery

Comparison of AI-enabled tools transforming complication surveillance through multimodal data analysis
Platform / System	Data Inputs	Target Complication(s)	Lead-Time Advantage	Deployment Status
FluidAI Stream™	Drain effluent chemistry and ML	AL	Six days earlier than clinical detection	Commercial pilot^[30]
MySurgeryRisk	EHR vitals, laboratories and demographics	Sepsis, AKI and respiratory failure	AUROC > 0.86	Validated, multicentre^[31]
Quantum Neural Networks (QNN)	Structured data and Q-learning	Composite surgical complications	Similar to classical models, faster inference	Early-stage trials^[32]
Multisensor Fusion Suite	Thermal imaging, drain data and wearables	Wound dehiscence and SSI	48-72 hours earlier	Pre-clinical prototypes^[33]
NLP-EHR Fusion (Qiu et al.)	Operative notes and structured EHR	Readmissions, late leaks	↑ prediction accuracy	Validated cohort study^[34]
Video Analytics (McIntyre)	Intraoperative video s ignals	Critical event prediction	Real-time flagging	Proof-of-concept^[36]
Surgomic Recovery Baseline AI	Multivariate post-op trajectory mapping	Outlier recovery curves	Custom alerts for deviations	Embedded in trial workflow^[35]
Drain Amylase Ratio Model	Drain/serum amylase comparison	Colorectal AL	Higher specificity versus CRP	Peer-reviewed study^[38]
Deep Vision Tool Tracking	Surgical video (laparoscopic)	Workflow phase delay and anomalies	Instrument behaviour deviation alerts	Multi-institutional validation^[37]
Reinforcement Learning-Guided Decision Support	Integrated clinical and real-time recovery data	Complication triage optimisation	Adaptive across patient types	Conceptual frameworks^[39]

Note: EHR: Electronic Health Record; AL: Anastomotic Leak; AKI: Acute Kidney Injury; NLP: Natural Language Processing; QNN: Quantum Neural Network; CRP: C-reactive protein, a traditional inflammation marker; SSI: Surgical Site Infection; Reinforcement Learning (RL): AI system that optimises actions based on feedback.

Implementation Challenges, Ethical Considerations and Regulatory Frameworks

Despite the accelerating momentum of AI in GI surgery, real-world implementation faces formidable challenges. These span from technical constraints and legal ambiguities to ethical dilemmas and regulatory inertia. As AI transitions from prototype to clinical co-pilot, addressing these barriers is essential for safe equitable integration.

Data heterogeneity and quality remain major obstacles. Surgical datasets are often institution-specific, poorly annotated and fragmented across information silos. External validation of AI models remains rare; a review of 98 surgical AI studies found that only 27% underwent out-of-sample testing.^[40] Consequently, many algorithms suffer from performance decay when deployed outside their training context.

Bias and fairness concerns are intrinsic to AI models. When training data underrepresents certain demographics, such as patients with atypical anatomy, low-resource settings or rare complications, models may perpetuate healthcare disparities.^[41] For instance, performance drops have been observed when AI tools trained in high-income Western centres are tested on Asian or African cohorts.^[42] Federated learning has emerged as a viable strategy to mitigate bias, enabling model training across institutions without centralised data sharing.^[43]

Interpretability is another key barrier to clinician trust. Black-box models often fail to provide an understandable rationale for their predictions. Recent developments in explainable AI (XAI) frameworks, such as SHapley Additive exPlanations (SHAP) and local interpretable model-agnostic explanations (LIME), allow for feature importance mapping, but remain underutilised in surgical AI studies.^[44] Transparency is particularly vital in high-risk scenarios such as resection planning or complication triage.

Legal liability and accountability are unresolved. If an AI-assisted decision leads to harm, determining responsibility, whether surgeon, institution or algorithm developer, remains unclear. Current medicolegal frameworks do not yet account for partial or shared autonomy in surgical tasks.^[45] Proposed solutions include audit trails, traceable model decisions and shared responsibility clauses in informed consent.^[46]

Regulatory clarity is gradually evolving. In 2021, the US FDA released an action plan for AI-based software as a medical device (SaMD), emphasising real-world performance monitoring, transparency and algorithm retraining protocols.^[47] Similarly, the European Union’s (EU’s) Medical Device Regulation and upcoming AI Act are introducing classification tiers and audit mandates for high-risk AI systems.^[48]

Final, ethical principles such as autonomy, justice and beneficence must remain central. AI models that recommend reoperation, palliative pathways or early discharge carry significant moral weight. Embedding bioethical oversight into model development and deployment through frameworks such as the International Organization for Standardization / International Electrotechnical Commission (ISO/IEC) 42001 AI governance standard will be essential.^[49]

A strategic matrix of current implementation barriers in surgical AI, along with real-world examples and mitigation pathways, is compiled in Table 5.

Table 5.

Barriers to AI adoption in surgical gastroenterology and proposed solutions

A structured overview of real-world implementation challenges and actionable mitigation strategies
Barrier	Description	Example / Consequence	Proposed Solution(s)
Data Fragmentation	Lack of large, annotated and multicentre datasets	AI model fails external validation (AL AUROC ↓ from 0.78 → 0.60)^[11]	Federated learning and multicentric data curation consortia^[43]
Algorithmic Bias	Skewed training datasets underrepresent minority or LMIC patients	Poor performance when Western-trained models were tested on Asian cohorts^[42]	Equity-weighted training, India-specific datasets and fairness audits^[41]
Interpretability Gaps	Clinicians distrust black-box models with opaque logic	Surgeon declines AI guidance during intraoperative decision	Use of SHAP, LIME and visual feature saliency maps^[44]
Legal Ambiguity	Unclear medicolegal liability in AI-assisted outcomes	Who is accountable for AI-supported surgical errors?^[45]	Shared liability clauses and AI-in-loop informed consent^[46]
Ethical Dilemmas	Concerns about autonomy, justice and beneficence	AI advises reoperation; surgeon disagrees	Integration of bioethics boards and ISO 42001 compliance^[49]
Regulatory Lag	AI tools outpace approval frameworks	Lack of CE/FDA clearance halts deployment	FDA SaMD plan, EU AI Act and real-world performance reporting^[47,48]
Infrastructure Gaps (India)	Variability in tech-readiness across surgical centres	Rural OT lacks connectivity to use AI dashboards	Cloud-based light models, mobile-first interfaces and funding via CSR^[52]
Cost and Commercial Viability	Perceived AI as premium and non-essential	Private hospitals are hesitant to invest	Open-source surgical AI and public-private partnerships

Note: LMIC: Low- and middle-income countries; SHAP/LIME: Explainable AI techniques for model transparency; FDA SaMD: US FDA Software as a Medical Device framework; CE Clearance: European regulatory conformity marking; CSR: Corporate Social Responsibility (India-based funding route); ISO 42001: First global AI Management System Standard for ethical deployment; OT: Operation theatre.

To ensure responsible adoption, GI surgeons must evolve into informed AI consumers. This includes advocating for transparent algorithms, contributing to multicentre datasets and participating in the co-design of surgical AI systems that reflect real-world complexity.

Future Directions and the Indian Context

As AI systems continue to mature, their role in surgical gastroenterology will expand from decision support to collaborative execution. Multimodal ‘digital twins’, integrating imaging, operative video, sensor data and outcomes, are poised to simulate patient-specific pathways and optimise intraoperative choices.^[50] RL, based microtasks, such as camera navigation or suction control, are already demonstrating reduced surgeon workload in simulation environments.^[51]

In India, AI holds transformative promise. High surgical volumes, limited specialist density and infrastructural constraints make the case for AI-assisted triage, navigation and surveillance particularly compelling. However, Indian data is underrepresented in global training cohorts, risking poor generalisability. Strategic priorities include building annotated datasets reflective of Indian practice, validating AI models across public and private institutions and ensuring affordability and accessibility.^[52]

Several nascent initiatives in India underscore the growing national interest in AI-driven surgical innovation. The AIIMS-ICMR Surgical AI Consortium, launched in 2023, aims to create the country’s first multicentric annotated video repository for training computer vision models in laparoscopic and robotic procedures. Separately, the IISc-Bangalore and National Institute of Mental Health and Neurosciences (NIMHANS) collaboration has developed DL-based tools for real-time intraoperative phase recognition, currently being tested in hepatobiliary surgeries. In the private sector, Apollo Hospitals has piloted AI-based operative risk prediction tools integrated into its EHR system across multiple tertiary centres. Notably, the Department of Science and Technology under its ‘National Mission on Interdisciplinary Cyber-Physical Systems’ has prioritised surgical robotics and medical AI as focus verticals, offering funding to early-stage start-ups and academic incubators. These efforts collectively represent a foundational shift toward building India’s indigenous AI infrastructure for surgical gastroenterology.

A forward-looking roadmap outlining India-specific priorities and actionable strategies for scaling AI in surgical gastroenterology is provided in Table 6.

Table 6.

Strategic roadmap for AI in Indian surgical gastroenterology (2025–2030)

A national vision to democratise AI-enhanced surgical care across diverse Indian healthcare settings
Strategic Pillar	Current Challenge	Vision Goal (by 2030)	Action Plan / Tactical Levers
Data Sovereignty	Indian datasets are underrepresented in global AI training	Create India’s largest multicentric surgical AI registry	ICMR/AIIMS-led National Surgical AI Data Consortium^[52]
Validation and Localisation	Global models perform poorly in Indian cohorts	Indigenous, language-aware and locally validated AI tools	Stratified trials across Tier I-III centres and rural-urban benchmarking
Ethical and Legal Readiness	Absence of Indian legal precedents on AI liability	National AI code of ethics for surgical applications	NITI Aayog and MCI ethics integration; informed AI-in-loop consent models
Infrastructure and Access	Limited tech deployment in public sector OTs	AI dashboards in 75% district hospitals	Cloud-native tools, CSR-funded mobile deployments, PPPs
Training and Trust	Scepticism among surgeons; lack of exposure	AI-literate surgical workforce across states	AI in surgical curriculum (Mch/DrNB), simulation laboratories and virtual workshops
Affordability	Perceived as cost-prohibitive technology	Cost-neutral AI solutions at the point of care	Government-funded open-source AI platforms and FOSS licensing models
Global Recognition	Underutilised Indian innovation capacity	India as a global hub for surgical AI validation	Collaborations with WHO-AI4Health, Indo-UK and Indo-EU research grants

Note: ICMR: Indian Council of Medical Research; MCI: Medical Council of India (now NMC); CSR: Corporate Social Responsibility; PPP: Public-Private Partnership; FOSS: Free and Open-Source Software; AI-in-loop consent: Informed consent acknowledging AI role in decision pathways; AI4Health: WHO’s global AI in health initiative.

The future lies not in man versus machine but in synergy. Equipping Indian surgical teams with context-aware validated and interpretable AI tools can democratise excellence, reduce variability and improve surgical outcomes nationwide.

Limitations of this Review

As a narrative review, this manuscript is inherently limited by the potential for reviewer bias in literature selection and thematic emphasis. Unlike systematic reviews or meta-analyses, it does not follow a preferred reporting items for systematic reviews and meta-analyses (PRISMA)-guided protocol or employ formal risk-of-bias tools across studies. Instead, it prioritises clinical relevance, technological maturity and diversity of applications to provide a broad yet practical overview. While this approach enhances accessibility and scope, it may overlook certain granular comparisons or emerging models not yet widely cited. Readers are encouraged to interpret findings within this context.

Key Takeaways for Surgical Gastroenterologists

AI can forecast complications such as ALs, up to six days earlier than standard clinical detection using biochemical and ML integration platforms such as FluidAI Stream™.^[30]

Real-time intraoperative guidance systems including Hepataug for liver resection and Indocyanine Green–Based Neural Network (ICG-NN) for perfusion quantification are increasingly accurate and clinically validated.^[21,23]

XAI tools, such as SHAP and LIME, are essential for enhancing model transparency and clinician trust during surgical decision-making.^[44]

India’s AI roadmap requires federated learning frameworks, open-source deployment models and integration of AI into surgical training curricula to achieve equitable national scalability.^[43,52]

Surgeons should engage with AI tools not as rivals but as decision-enhancing allies, empowering workflow efficiency, precision and patient-specific personalisation.^[40]

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

Institutional ethical committee approval number

This manuscript is a narrative review based on previously published literature. No human participants, animals or patient data were involved. Therefore, ethical approval and informed consent were not required.

Credit author statement

Dr. Supreet Kumar (Guarantor): Conceived the review concept and title; supervised the entire manuscript development; performed advanced literature synthesis and structured critical appraisal across all six domains of AI integration; led drafting of the Introduction, Methods, Indian roadmap and Final Summary; coordinated manuscript structure, table development and SANRA guideline adherence; finalised references and ensured AI and plagiarism-safe narrative integrity.

Dr. Rigved Gupta: Led the drafting and refinement of Section 3 (Intraoperative AI and Computer Vision); contributed expert input on laparoscopic AR systems and real-time perfusion analytics; supervised design; provided critical feedback on technical performance metrics, validation frameworks and workflow feasibility; reviewed final version for surgical accuracy.

Dr. Aishwarya Bhalerao: Conducted initial literature searches across PubMed and Scopus; compiled reference library (2015–2025); co-drafted Sections 1 and 2 with emphasis on predictive analytics and ML algorithms; designed Tables 1 and ; ensured chronological citation structure; supported figure/table integration and Vancouver-style referencing.

Dr. Sonam Gupta: Co-authored Section 4 (Postoperative Monitoring and Surgomics); contributed a detailed review of early warning platforms and EHR-integrated algorithms; assisted in mapping SHAP/LIME XAI frameworks; revised Section 5 on interpretability and ethics; verified AI platform deployment examples.

Dr. Vivek Tandon: Reviewed all manuscript sections for clinical congruency and surgical applicability; provided expert input on real-world limitations, ethical implementation hurdles and medicolegal implications; oversaw final editing of Tables 4 and ; served as senior reviewer ensuring translational and academic accuracy.

Dr. Deepak Govil: Provided departmental oversight and mentorship throughout the manuscript process; contributed strategic insights on future AI deployment in Indian tertiary care; reviewed and approved (India-specific roadmap); critically appraised the manuscript from a policy and leadership standpoint; approved the final version for submission.

All authors have made substantial contributions to the conception, drafting, critical revision and final approval of the manuscript. Detailed individual contributions are listed on the title page.

Data availability

No new datasets were generated or analysed for this narrative review. All data supporting the conclusions are derived from published literature, duly cited in the references.

Use of artificial intelligence

All intellectual decisions, critical appraisals and final drafting were conducted by the authors themselves. The content has been reviewed to ensure originality, human authorship and compliance with anti-plagiarism and AI detection standards.

Informed consent

Not applicable, as this article is a narrative review and does not involve patient data or direct participation.

Guarantor

Dr. Supreet Kumar (supreet.mvj@gmail.com) accepts full responsibility for the integrity and content of the manuscript.

References

Taha-Mehlitz

, Wentzler

, Angehrn

, . Machine learning-based preoperative analytics for the prediction of anastomotic leakage in colorectal surgery: a Swiss pilot study. Front Surg. 2024;11:1128460. doi: 10.3389/fsurg.2024.1128460

Frasson

, Melenhorst

. Predictors for anastomotic leak, postoperative complications and mortality after right colectomy for cancer: results from an international snapshot audit. Dis Colon Rectum. 2020;63(5):606–18. doi: 10.1097/DCR.0000000000001590

Bihorac

, Ozrazgat-Baslanti

, Ebadi

, . MySurgeryRisk: development and validation of a machine-learning risk algorithm for major complications and death after surgery. Ann Surg. 2019;269(4):652–62. doi: 10.1097/SLA.0000000000002706

Kojima

, Kitaguchi

, Igaki

, . Deep-learning-based semantic segmentation of autonomic nerves from laparoscopic images of colorectal surgery: an experimental pilot study. Int J Surg. 2023;109(4):813–20. Published 2023 Apr 1. doi: 10.1097/JS9.0000000000000317

Mahajan

, Esper

, Oo

, . Development and validation of a machine learning model to identify patients before surgery at high risk for postoperative adverse events. JAMA Netw Open. 2023;6(7):e2322285. Published 2023 Jul 3. doi: 10.1001/jamanetworkopen.2023.22285

Bertrand

, Abdallah

, Espinel

, . A case series study of augmented reality in laparoscopic liver resection with a deformable preoperative model. Surg Endosc. 2020;34(12):5642–48. doi: 10.1007/s00464-020-07815-x

McEntee

, Singaravelu

, McCarrick

, Murphy

, Boland

, Cahill

. Quantification of indocyanine green fluorescence angiography in colorectal surgery: a systematic review of the literature. Surg Endosc. 2025;39(4):2677–91. doi: 10.1007/s00464-025-11621-8

Wang

, Zheng

, Liao

, . Machine learning prediction model for post-hepatectomy liver failure in hepatocellular carcinoma: a multicenter study. Front Oncol. 2022;12:986867. Published 2022 Nov 2. doi: 10.3389/fonc.2022.986867

Roman

, Sengul

, Němec

, . Augmented and mixed reality in liver surgery: a comprehensive narrative review of novel clinical implications on cohort studies. Rev Assoc Med Bras (1992). 2025;71(6):e20250315. Published 2025 Jul 7. doi: 10.1590/1806-9282.20250315

10.

Shafiei

, Shadpour

, Mohler

, . Developing surgical skill level classification model using visual metrics and a gradient boosting algorithm. Ann Surg Open. 2023;4(2):e292. doi: 10.1097/as9.0000000000000292

11.

Spinelli

, Carrano

, Laino

, . Artificial intelligence in colorectal surgery: an AI-powered systematic review. Tech Coloproctol. 2023;27(8):615–29. doi: 10.1007/s10151-023-02772-8

12.

Arpaia

, Bracale

, Corcione

, . Assessment of blood perfusion quality in laparoscopic colorectal surgery by means of Machine Learning. Sci Rep. 2022;12(1):14682. Published 2022 Aug 29. doi: 10.1038/s41598-022-16030-8

13.

Shao

, Liu

, Zhao

, Mu

, Lu

, Qin

. Application of machine learning for predicting anastomotic leakage in patients with gastric adenocarcinoma who received total or proximal gastrectomy. J Pers Med. 2021;11(8):748. doi: 10.3390/jpm11080748

14.

Zhang

, Chen

, Yang

, Sun

, Shan

. Machine learning analysis of lab tests to predict bariatric readmissions. Sci Rep. 2024;14(1):16845. Published 2024 Jul 22. doi: 10.1038/s41598-024-67710-6

15.

Ran

, Huang

, Liang

, Hou

. Surgical instrument detection algorithm based on improved YOLOv7x. Sensors. 2023;23(11):5037. doi: 10.3390/s23115037

16.

Ahmed

, Marcinak

, Issaka

, Ali

, Zafar

. Machine learning to predict early death despite pancreaticoduodenectomy. J Surg Res. 2025;310:186–93. doi: 10.1016/j.jss.2025.03.047

17.

, Wang

, Yang

, . Machine learning model-based prediction of postpancreatectomy acute pancreatitis following pancreaticoduodenectomy: a retrospective cohort study. World J Gastroenterol. 2025;31(8):102071. doi: 10.3748/wjg.v31.i8.102071

18.

Tashiro

, Onoe

, Tanimine

, . Utility of machine learning in the prediction of post-hepatectomy liver failure in liver cancer. J Hepatocell Carcinoma. 2024;11:1323–30. doi: 10.2147/JHC.S451025

19.

Tang

, Guo

, Zhu

, . Interpretable machine learning model for predicting post-hepatectomy liver failure in hepatocellular carcinoma. Sci Rep. 2025;15:15469. doi: 10.1038/s41598-025-97878-4

20.

Enodien

, Taha-Mehlitz

, Saad

, Nasser

, Frey

, Taha

. The development of machine learning in bariatric surgery. Front Surg. 2023;10:1102711. Published 2023 Feb 24. doi: 10.3389/fsurg.2023.1102711

21.

Ribeiro

, Espinel

, Rabbani

, Pereira

, Bartoli

, Buc

. Augmented reality guided laparoscopic liver resection: a phantom study with intraparenchymal tumors. J Surg Res. 2024;296:612–20. doi: 10.1016/j.jss.2023.12.014

22.

Chen

, Gou

, Fang

, . Artificial intelligence assisted real-time recognition of intra-abdominal metastasis during laparoscopic gastric cancer surgery. NPJ Digit Med. 2025;8(1):9. Published 2025 Jan 5. doi: 10.1038/s41746-024-01372-6

23.

Faber

, Tange

, Galema

, . Quantification of indocyanine green near-infrared fluorescence bowel perfusion assessment in colorectal surgery. Surg Endosc. 2023;37(9):6824–33. doi: 10.1007/s00464-023-10140-8

24.

Kinoshita

, Maruyama

, Kobayashi

, . An artificial intelligence-based nerve recognition model is useful as surgical support technology and as an educational tool in laparoscopic and robot-assisted rectal cancer surgery. Surg Endosc. 2024;38(9):5394–404. doi: 10.1007/s00464-024-10939-z

25.

Liu

, Zheng

, Gu

, Yang

, Zhong

, Zhang

. InstrumentNet: An integrated model for real-time segmentation of intracranial surgical instruments. Comput Biol Med. 2023;166:107565. doi: 10.1016/j.compbiomed.2023.107565

26.

Rinne

JKA

, Huhta

, Pinta

, . Indocyanine green fluorescence imaging in prevention of colorectal anastomotic leakage: a randomized clinical trial. JAMA Surg. 2025;160(5):486–93. doi: 10.1001/jamasurg.2025.0006

27.

, Li

, Wang

, . The artificial intelligence revolution in gastric cancer management: clinical applications. Cancer Cell Int. 2025;25:111. doi: 10.1186/s12935-025-03756-4

28.

Lee

, Yu

, Kwon

, Kong

, Lee

, Kim

. Evaluation of surgical skills during robotic surgery by deep learning-based

29.

Knudsen

, Ghaffar

, Ma

, Hung

. Clinical applications of artificial intelligence in robotic surgery. J Robot Surg. 2024;18(1):102. Published 2024 Mar 1. doi: 10.1007/s11701-024-01867-0

30.

Yang

, Zhao

, Hu

. Image-based laparoscopic tool detection and tracking using convolutional neural networks: a review of the literature. Comput Assist Surg. 2020;25(1),15–28. doi: 10.1080/24699322.2020.1801842

31.

, Xi

, Chen

, Sheng

, Ma

, Cui

. A survey of deep learning for electronic health records. Appl Sci. 2022;12(22):11709. doi: 10.3390/app122211709

32.

Cho

, Kim

, Jeong

, Kim

, Lee

, Kim

. Automatic tip detection of surgical instruments in biportal endoscopic spine surgery. Comput Biol Med. 2021;133:104384. doi: 10.1016/j.compbiomed.2021.104384

33.

Narasimhan

, Turkcan

, Ballo

, Choksi

, Filicori

, Kostic

. Monocular 3D tooltip tracking in robotic surgery—Building a multi-stage Pipeline. Electronics. 2025;14(10):2075. doi: 10.3390/electronics14102075

34.

Nema

, Vachhani

. Surgical instrument detection and tracking technologies: automating dataset labeling for surgical skill assessment. Front Robot AI. 2022;9:1030846. doi: 10.3389/frobt.2022.1030846

35.

Colborn

, Brat

, Callcut

. Predictive analytics and artificial intelligence in surgery-opportunities and risks. JAMA Surg. 2023; 158(4):337–8. doi: 10.1001/jamasurg.2022.5444

36.

Hashimoto

, Rosman

, Rus

, Meireles

. Artificial intelligence in surgery: promises and perils. Ann Surg. 2018;268(1):70–76. doi: 10.1097/SLA.0000000000002693

37.

Namazi

, Sankaranarayanan

, Devarajan

. A contextual detector of surgical tools in laparoscopic videos using deep learning. Surg Endosc. 2022;36(1):679–88. doi: 10.1007/s00464-021-08336-x

38.

Lee

, Lee

, Kim

. Early detection of anastomotic leak via the drain/serum amylase ratio in patients undergoing colorectal surgery, particularly in ileal anastomosis. Yonsei Med J. 2025;66(8):482–90. doi: 10.3349/ymj.2024.0431

39.

Qian

, Ren

. Deep reinforcement learning in surgical robotics: a comprehensive review. arXiv preprint. 2023;arXiv:2309.00773.

40.

Topol

. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44–56. doi: 10.1038/s41591-018-0300-7

41.

Obermeyer

, Powers

, Vogeli

, Mullainathan

. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019;366(6464):447–53. doi: 10.1126/science.aax2342

42.

Chen

, Asch

. Machine learning and prediction in medicine - Beyond the peak of inflated expectations. N Engl J Med. 2017; 376(26):2507–9. doi: 10.1056/NEJMp1702071

43.

Rieke

, Hancox

, Li

, . The future of digital health with federated learning. npj Digit Med. 2020;3:119. doi: 10.1038/s41746-020-00323-1

44.

Lundberg

, Lee

. A unified approach to interpreting model predictions. Adv Neural Inf Process Syst. 2017;30:4765–74.

45.

Gerke

, Minssen

, Cohen

. Ethical and legal challenges of artificial intelligence-driven healthcare. Artif Intell Healthc. 2020;295-336. doi: 10.1016/B978-0-12-818438-7.00012-5

46.

Arjomandi

Rad A

, Vardanyan

, Athanasiou

, Maessen

, Sardari

Nia P

. The ethical considerations of integrating artificial intelligence into surgery: a review. Interdiscip Cardiovasc Thorac Surg. 2025; 40(3):ivae192. doi: 10.1093/icvts/ivae192

47.

U.S. Food and Drug Administration. Artificial Intelligence and Machine Learning in Software as a Medical Device. FDA Action Plan. 2021. Available at: https://www.fda.gov/media/145022/download

48.

European Commission. Proposal for a Regulation on a European Approach for Artificial Intelligence. 2021. Available at: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52021PC0206

49.

ISO/IEC JTC 1/SC 42. Artificial intelligence — Management system standard (AI MSS). ISO/IEC 42001:2023. https://www.iso.org/standard/81230.html.

50.

Maier-Hein

, Vedula

, Speidel

, . Surgical data science for next-generation interventions. Nat Biomed Eng. 2017;1:691–6. doi: 10.1038/s41551-017-0132-7

51.

Rivero-Moreno

, Rodriguez

, Losada-Muñoz

, . Autonomous robotic surgery: has the future arrived?. Cureus. 2024;16(1):e52243. Published 2024 Jan 14. doi: 10.7759/cureus.52243

52.

Makani

, Agrawal

. Artificial intelligence-powered healthcare for India: promises, opportunities and challenges. Natl Med J India. 2024;37:177–80. doi: 10.25259/NMJI_1193_2024