Artificial intelligence in expanding hepatic resection boundaries: Integrating portal flow modulation and regenerative strategies

Abstract

Post-hepatectomy liver failure (PHLF) remains the principal barrier to major hepatectomy despite its curative potential for hepatocellular carcinoma and colorectal liver metastases; accordingly, this narrative review (aligned with PRISMA guidance) synthesizes evidence from PubMed, Scopus, Web of Science, and Google Scholar (January 2010 to October 2025) on how artificial intelligence (AI)-spanning automated liver and vascular segmentation, volumetry, radiomics, risk prediction, and virtual planning-can be integrated with portal flow modulation and regenerative strategies [portal vein embolization (PVE), liver venous deprivation (LVD), and associating liver partition and portal vein ligation for staged hepatectomy (ALPPS)] to safely expand resection boundaries. Across English-language studies meeting predefined inclusion criteria, AI demonstrated high segmentation accuracy (dice > 0.95) and robust PHLF prediction (area under the curve ≈ 0.82-0.94), while patient-specific 3D/VR models altered operative plans in approximately 44% of major hepatectomies to improve future liver remnant (FLR) preservation. Stepwise hypertrophy outcomes favored PVE (approximately 37%-40% in 3-6 weeks) and LVD (approximately 50%-70% in 2-4 weeks), with ALPPS achieving rapid hypertrophy (approximately 60%-80% in 7-10 days) at the cost of higher morbidity and mortality, supporting selective use. Importantly, radiomics and magnetic resonance imaging-derived features refined functional FLR assessment beyond volume alone and enhanced individualized risk stratification. Overall, pairing AI-enabled planning and outcome prediction with targeted hypertrophy strategies can broaden indications for curative liver resection while mitigating PHLF risk, though prospective validation, bias mitigation, workflow integration, and cost-effectiveness analyses are required before routine adoption.

Key Words: Hepatic resection; Artificial intelligence; Portal vein embolization; Liver venous deprivation; Associating liver partition and portal vein ligation for staged hepatectomy

Core Tip: Advanced artificial intelligence tools-automated segmentation, radiomics, and explainable risk models-optimize future liver remnant assessment and surgical planning. When combined with portal flow modulation (portal vein embolization/liver venous deprivation) and judicious associating liver partition and portal vein ligation for staged hepatectomy use, they offer a pragmatic pathway to expand resection candidacy without compromising safety, provided implementation is guided by rigorous validation and ethical safeguards.

INTRODUCTION

Hepatic malignancies, including hepatocellular carcinoma and colorectal liver metastases (CRLM), represent a substantial global health burden. Surgical resection remains the cornerstone of curative treatment, offering the best long-term survival outcomes[1]. However, the ability to perform major hepatectomy is often limited by the risk of post-hepatectomy liver failure (PHLF), which occurs when the remaining liver parenchyma is insufficient to maintain hepatic function. This complication carries mortality rates ranging from 8%-12% and significantly impacts patient outcomes[2].

The adequacy of the future liver remnant (FLR) is determined by multiple factors including liver volume, functional capacity, and underlying parenchymal quality[3]. Traditional assessment methods, while valuable, often fail to capture the complex interplay of anatomical, functional, and pathological factors that determine post-resection outcomes. This limitation has historically prevented many patients with extensive liver disease from undergoing potentially curative resections[4].

To address insufficient FLR, several portal flow modulation techniques have been developed to induce contralateral liver hypertrophy. Portal vein embolization (PVE), first described by Makuuchi in 1990, selectively occludes portal venous branches to redirect blood flow, inducing compensatory hypertrophy in the non-embolized liver[5]. PVE typically achieves mean FLR hypertrophy of 37%-40% within 3-6 weeks. However, PVE alone may produce insufficient hypertrophy in 5%-10% of patients, leading to procedure dropout[6].

Advanced techniques including liver venous deprivation (LVD), which combines portal and hepatic vein embolization, have demonstrated superior hypertrophy responses. LVD achieves higher FLR volume increases (up to 68%-70%) and faster kinetic growth rates compared to PVE alone[7]. The associating liver partition and portal vein ligation for staged hepatectomy (ALPPS) procedure represents an even more aggressive approach, inducing rapid hypertrophy (median 74% in 9 days) through parenchymal transection combined with portal ligation. However, ALPPS is associated with higher complication rates (44%) and mortality (12%) compared to conventional approaches[8].

The integration of artificial intelligence (AI) and machine learning (ML) into hepatic surgery represents a paradigm shift in preoperative planning, risk stratification, and outcome prediction. AI-driven technologies offer several advantages including automated liver volumetry, tumor segmentation, vascular reconstruction, and predictive modeling. Deep learning algorithms, particularly convolutional neural networks (CNNs), have achieved remarkable accuracy in automatic liver segmentation from computed tomography (CT) and magnetic resonance imaging (MRI) images, with dice scores exceeding 95%[9].

ML models can predict PHLF with area under the curve (AUC) values of 0.82-0.94, outperforming traditional scoring systems. AI-assisted 3D modeling changes operative planning in up to 44% of cases, leading to optimized FLR preservation. Radiomics analysis extracts quantitative features from medical images that correlate with histopathological outcomes and recurrence patterns. These technologies enable personalized surgical approaches tailored to individual patient characteristics[10].

This narrative review aims to: (1) Synthesize current evidence on AI applications in liver surgery, focusing on volumetric assessment, segmentation, and outcome prediction; (2) Examine portal flow modulation techniques and their mechanisms of inducing liver hypertrophy; (3) Explore molecular pathways of liver regeneration; (4) discuss the integration of AI with regenerative strategies to expand resection boundaries; and (5) Address current limitations, ethical considerations, and future directions in this rapidly evolving field.

METHODOLOGY

Search strategy and selection criteria

This narrative review was conducted following PRISMA guidelines adapted for narrative reviews[11,12]. A comprehensive literature search was performed across multiple databases including PubMed, Scopus, Web of Science, and Google Scholar. The search strategy combined terms related to AI, ML, hepatic resection, PVE, liver regeneration, and predictive modeling.

Search terms included: “Artificial intelligence” OR “machine learning” OR “deep learning” AND “hepatic resection” OR “liver surgery” OR “hepatectomy” AND “portal vein embolization” OR “ALPPS” OR “liver venous deprivation” AND “liver regeneration” OR “future liver remnant” OR “post-hepatectomy liver failure”. Additional searches focused on specific AI applications including “radiomics”, “liver segmentation”, “volumetry”, and “predictive models”.

The search was limited to English-language publications from January 2010 to October 2025, encompassing the period of significant AI advancement in medical imaging and surgical applications. Manual searches of reference lists from key articles were performed to identify additional relevant studies. Both peer-reviewed original research and review articles were included to provide comprehensive coverage of this multidisciplinary topic.

Inclusion and exclusion criteria

Studies were included if they: (1) Addressed AI or ML applications in liver surgery, imaging, or outcome prediction; (2) Described portal flow modulation techniques including PVE, LVD, or ALPPS; (3) Investigated molecular mechanisms of liver regeneration; (4) Examined predictive models for PHLF or surgical outcomes; and (5) Provided original data or systematic reviews on relevant topics. Exclusion criteria included: (1) Non-English publications; (2) Case reports with fewer than 5 patients (except for novel techniques); (3) Studies without clear methodology; and (4) Publications outside the defined time period. Figure 1 demonstrates the PRISMA Flow diagram of the current narrative review.

Figure 1 PRISMA 2009 flow diagram.

Data extraction and synthesis

Given the narrative nature of this review and the heterogeneity of included studies, formal meta-analysis was not performed. Instead, data were synthesized thematically into major categories: (1) AI applications in preoperative planning; (2) Portal flow modulation techniques; (3) Liver regeneration mechanisms; (4) Outcome prediction and risk stratification; and (5) Integration strategies and clinical implementation. Key findings, study methodologies, and clinical implications were extracted and synthesized to provide a comprehensive overview of the current state of knowledge.

AI IN PREOPERATIVE PLANNING AND ASSESSMENT

Automated liver segmentation and volumetry

Accurate volumetric assessment of the FLR is critical for predicting post-hepatectomy outcomes. Traditional manual segmentation is time-consuming, operator-dependent, and subject to inter-observer variability[13]. Deep learning-based automated segmentation has revolutionized this process, achieving segmentation accuracy comparable to expert radiologists while reducing processing time from hours to minutes[9].

Multiple CNN architectures have been developed for liver segmentation, with U-Net and its variants demonstrating superior performance (Figure 2). A study by Wu et al[14] implemented a 2D U-Net achieving dice similarity coefficients of 95.8% for liver segmentation on CT images. The medical open network for AI framework has facilitated the development of standardized, reproducible segmentation models. Three-dimensional segmentation approaches that leverage volumetric context have further improved accuracy, particularly for challenging anatomical variants[14].

Figure 2 The workflow related to the proposed method for explainable liver segmentation.

Automated hepatic vessel segmentation represents another critical application, enabling precise surgical planning and vascular risk assessment[15]. Deep learning algorithms can accurately segment portal veins, hepatic veins, and hepatic arteries on portal venous phase CT, achieving mean intersection over union values exceeding 0.86. This capability allows for virtual hepatectomy simulation and assessment of vascular involvement prior to surgery[16].

AI-assisted volumetry extends beyond simple volume calculation to functional assessment. Integration of gadoxetic acid-enhanced MRI features enables prediction of functional liver capacity, which may better correlate with PHLF risk than volume alone. Radiomics features extracted from the hepatobiliary phase can predict inadequate FLR function with AUC values of 0.84[17]. Table 1 summarizes the AI applications in liver surgery.

Table 1 Artificial intelligence and machine learning applications in liver surgery.

Application area	AI/ML model type	Data source	Clinical impact	Performance metric
Liver segmentation/volumetry	CNN (U-Net, variants)	CT, MRI	Automated FLR measurement, resection planning	Dice coefficient > 0.95
FLR function prediction	Radiomics, ML classifiers	MRI (Gd-EOB-DTPA), CT	Predicts PHLF and functional margins	AUC 0.82-0.94
Outcome prediction (PHLF, complications)	Gradient boosting, Light GBM	EHR, imaging	Individualized risk, clinical DSS	AUC 0.82-0.94
Tumor segmentation/classification	Deep CNN	CT, MRI	Automated detection, margin planning	Accuracy > 93%
Intraoperative decision support	Explainable ML, AR	Video, segmentation	Real-time guidance, workflow efficiency	Not routinely quantified

CNN: Convolutional neural network; FLR: Future liver remnant; EOB-DTPA: Gadoxetic acid (gadoxetate disodium; ethoxybenzyl-diethylenetriamine pentaacetic acid); PHLF: Post-hepatectomy liver failure; DSS: Decision support system; AUC: Area under the curve; CT: Computed tomography; AI: Artificial intelligence; ML: Machine learning; MRI: Magnetic resonance imaging; AR: Augmented reality.

3D reconstruction and virtual surgical planning

Patient-specific 3D liver models generated from preoperative imaging provide surgeons with unprecedented visualization of hepatic anatomy. AI-assisted 3D reconstruction software can automatically identify and label liver segments, tumors, and vascular structures according to anatomical classifications. These models enable virtual resection planning, allowing surgeons to simulate different operative approaches and optimize the preservation of functional parenchyma[18].

A multicenter study by Sanjeevi et al[19] demonstrated that AI-generated 3D models changed surgical planning in 44% of major hepatectomy cases, leading to increased FLR preservation. The technology proved particularly valuable for complex parenchyma-sparing resections where precise anatomical understanding is critical. Standardized FLR calculations based on these models showed better correlation with postoperative outcomes compared to conventional 2D measurements[20].

Virtual reality and augmented reality (AR) technologies integrated with AI-segmented models offer immersive preoperative planning experiences. Surgeons can visualize tumor locations, vascular relationships, and resection planes in three dimensions, enhancing spatial understanding and potentially improving surgical precision. Intraoperative AR navigation systems overlay preoperative planning data onto the surgical field, providing real-time guidance during complex dissections[21-23].

Complexity scoring and risk stratification

Liver resection complexity varies substantially based on tumor location, proximity to major vessels, extent of resection, and patient factors. AI models have been developed to automatically assess surgical complexity from preoperative imaging, providing objective risk stratification. A deep learning model by Ali et al[24] predicted intraoperative liver resection complexity with 79.4% accuracy, outperforming individual surgeon predictions.

The hepatic central zone (HCZ) framework, defined by the confluence of major hepatic vessels, provides an anatomical reference for complexity assessment. Tumors located within or adjacent to the HCZ require more technically demanding resections with higher complication risks. AI algorithms can automatically identify the HCZ from CT scans and quantify tumor relationships to this critical zone, enabling standardized complexity grading[25,26].

ML models incorporating multiple preoperative variables predict specific complications including bile leakage, hemorrhage, and liver failure. An explainable AI model using SHapley Additive exPlanations (SHAP) identified key predictors of outcomes in laparoscopic liver surgery for segments 7 and 8, providing interpretable risk assessments to guide surgical decision-making[27,28]. Figure 3 demonstrates AI integration in hepatic surgery workflow.

Figure 3 Artificial intelligence integration in hepatic surgery workflow.

PORTAL FLOW MODULATION TECHNIQUES FOR LIVER HYPERTROPHY

PVE: Mechanisms and outcomes

PVE induces FLR hypertrophy by redirecting portal blood flow from the deportalized lobe to the FLR, triggering compensatory regeneration. The procedure involves percutaneous transhepatic catheterization and selective embolization of portal vein branches corresponding to the planned resection territory. Various embolic materials including polyvinyl alcohol, N-butyl cyanoacrylate (NBCA), and coils have been employed, with NBCA demonstrating superior hypertrophy induction[29,30].

Meta-analyses of PVE outcomes show technical success rates of 99.3% with mean FLR hypertrophy of 37.9%. The median time to adequate hypertrophy is 3-6 weeks, with kinetic growth rates of approximately 2% per week[14,16]. However, 5%-10% of patients fail to achieve sufficient hypertrophy, and 2.8% demonstrate insufficient response to proceed with planned resection. Complications occur in less than 5% of cases, primarily including portal vein thrombosis, hepatic abscess, and pleural effusion[31,32].

Factors predicting hypertrophy response include baseline FLR volume, liver function, absence of chemotherapy-induced liver injury, and absence of severe fibrosis. Sarcopenia, defined by low skeletal muscle index, negatively correlates with hypertrophy degree, highlighting the importance of nutritional status. Additional segment 4 embolization enhances hypertrophy in extended right hepatectomy candidates, though at increased technical complexity[33,34].

The cellular and molecular mechanisms underlying PVE-induced hypertrophy involve a complex cascade of growth factors and cytokines. Portal flow deprivation triggers hepatocyte proliferation through activation of the interleukin-6 (IL-6)/STAT3 pathway and hepatocyte growth factor (HGF)/Met signaling. Concurrently, the deportalized liver undergoes atrophy through apoptosis and reduced protein synthesis[24,35].

LVD: Enhanced hypertrophy

LVD combines PVE with simultaneous or sequential hepatic vein embolization, achieving more pronounced FLR hypertrophy than PVE alone. The technique addresses a limitation of PVE: Intrahepatic collateral formation that may partially maintain portal perfusion to the embolized liver. By additionally occluding hepatic venous outflow, LVD creates complete vascular deprivation, maximizing regenerative stimulus to the FLR[36,37].

Comparative studies demonstrate superior volumetric and functional outcomes with LVD vs PVE. A systematic review by Deshayes et al[38] found LVD associated with higher FLR volume increases (mean difference 13.5%), faster kinetic growth rates, and lower failure-to-resect rates due to insufficient hypertrophy. Importantly, FLR functional capacity measured by hepatobiliary scintigraphy increases 1.5-1.7-fold more than volume, suggesting enhanced regenerative quality. Modified radiation lobectomy combining selective internal radiation therapy (SIRT) with portal flow modulation represents another approach to induce hypertrophy while providing tumor control[39,40].

Technical considerations for LVD include the approach to hepatic vein embolization. Traditional transhepatic percutaneous access carries risks of bleeding, pneumothorax, and injury to the FLR. Modified transvenous approaches accessing hepatic veins through the inferior vena cava avoid these risks while maintaining technical success rates near 100%. Segmentary analysis reveals that segments 1 and 2-3 demonstrate significantly greater hypertrophy after LVD compared to PVE, potentially improving resectability for centrally located tumors[41,42].

Complications of LVD parallel those of PVE with slightly higher rates of transient liver dysfunction. Post-procedural elevations in transaminases typically peak at 2-3 days and normalize within 1-2 weeks. Serious complications including PHLF occur in less than 3% of cases, with 90-day mortality rates of 0%-2%. Patient selection remains critical, with baseline liver function (Child-Pugh A) and adequate remnant liver quality being essential prerequisites[43,44].

ALPPS: Rapid hypertrophy with increased risk

The ALPPS procedure achieves remarkably rapid liver hypertrophy by combining portal vein ligation with parenchymal transection in the first stage, followed by removal of the deportalized liver 7-14 days later. This technique produces median FLR hypertrophy of 74% within 9 days, substantially faster than PVE (23 days). The ability to complete staged resection in 85%-95% of patients represents an advantage over PVE, where dropout rates of 19%-30% due to inadequate hypertrophy or tumor progression occur[45,46].

However, ALPPS carries significantly higher morbidity and mortality compared to conventional approaches. Major complications (Clavien-Dindo ≥ IIIb) occur in 44% of patients, with 90-day mortality rates of 12% in early series. Specific complications include postoperative liver failure (15%-20%), bile leakage (10%-15%), intra-abdominal infections, and wound complications. Factors associated with adverse outcomes include age ≥ 60 years, body mass index ≥ 30 kg/m², low postoperative albumin, and classical ALPPS vs partial variants[47,48].

Modified ALPPS techniques aim to reduce morbidity while preserving rapid hypertrophy. Partial ALPPS performs incomplete parenchymal transection or uses radiofrequency ablation to create an ischemic plane, achieving similar hypertrophy (65.2%) with lower complication rates (20%) compared to complete ALPPS. Minimally invasive approaches for stage 1, including laparoscopic or robotic portal ligation with partial transection, further reduce complications while maintaining efficacy[49,50].

The molecular mechanisms underlying ALPPS-induced hypertrophy differ from PVE through additional activation of regenerative pathways. Parenchymal transection triggers inflammatory cascades with release of damage-associated molecular patterns and cytokines including tumor necrosis factor-α (TNF-α) and IL-6. This creates a more potent regenerative stimulus than portal flow deprivation alone. The SPHK1/S1P pathway is distinctly activated in ALPPS, contributing to hepatocyte proliferation[51,52]. Table 2 summarize the portal flow modulation techniques.

Table 2 Comparison of portal flow modulation techniques.

Technique	Mechanism	Median FLR hypertrophy	Hypertrophy time	Main risks	Indications
Portal vein embolization (PVE)	Portal inflow occlusion	37%-40%	3-6 weeks	Low (abscess, thrombosis)	Insufficient FLR, most common bridge
Liver venous deprivation	PVE + hepatic vein embolization	50%-70%	2-4 weeks	Slightly higher dysfunction risk	Inadequate hypertrophy after PVE
Associating liver partition and portal vein ligation	Portal ligation + parenchymal transection	60%-80%	7-10 days	Higher: Infection, liver failure	Unresectable with rapid hypertrophy need

FLR: Future liver remnant.

MOLECULAR MECHANISMS OF LIVER REGENERATION

Cytokines and growth factors

Liver regeneration represents a complex orchestrated process involving multiple signaling pathways controlled by cytokines and growth factors. The regenerative response can be divided into three phases: Initiation (0-4 hours), proliferation (4-72 hours), and termination (72 hours onward). Each phase is characterized by specific molecular events and cellular responses[53,54].

The initiation phase is dominated by inflammatory cytokines, particularly TNF-α and IL-6, which prime hepatocytes to enter the cell cycle. TNF-α, secreted by Kupffer cells in response to portal flow changes or tissue injury, binds to TNF receptor 1 on hepatocytes, activating NF-κB signaling. This pathway induces immediate early genes including c-Fos, c-Jun, and c-Myc, transitioning quiescent hepatocytes from G0 to G1 phase. IL-6, also produced by Kupffer cells and stellate cells, activates the STAT3 pathway through gp130 receptor binding, promoting hepatocyte survival and proliferation[55,56].

The proliferation phase is characterized by mitogenic growth factors, principally HGF and epidermal growth factor (EGF). HGF, produced by hepatic stellate cells and released from extracellular matrix stores, binds to the Met receptor on hepatocytes, activating multiple downstream pathways including ERK/MAPK, PI3K/Akt, and mTOR. These cascades drive hepatocyte DNA synthesis and cell division. EGF and its receptor (EGFR) similarly promote proliferation through MAPK activation. Additional growth factors including vascular endothelial growth factor support angiogenesis necessary for regenerating parenchyma[57,58].

The termination phase involves anti-proliferative signals that limit regeneration once appropriate liver mass is restored. Transforming growth factor-β plays a central role in termination, inhibiting hepatocyte proliferation and promoting extracellular matrix remodeling. The Hippo/YAP signaling pathway acts as a mechanosensor, responding to tissue tension and contact inhibition to halt proliferation. Bile acids, whose concentrations increase during regeneration, activate the farnesoid X receptor (FXR) and TGR5, contributing to termination signals[54,55].

Regenerative pathways and cellular mechanisms

Beyond cytokine and growth factor signaling, multiple intracellular pathways regulate hepatocyte proliferation during regeneration. The Wnt/β-catenin pathway is essential for hepatocyte proliferation, with β-catenin nuclear translocation activating transcription of genes involved in cell cycle progression. The mTOR pathway, particularly mTORC1, integrates nutrient sensing with growth signals, driving protein synthesis and cell growth necessary for regeneration[58,59].

Nuclear receptors including peroxisome proliferator-activated receptors (PPARs) and liver X receptors modulate metabolic reprogramming during regeneration. Hepatocytes must balance energy-intensive proliferation with maintenance of critical metabolic functions. Autophagy provides amino acids and energy substrates to support regeneration while removing damaged organelles[60,61].

Non-coding RNAs, particularly microRNAs (miRNAs), have emerged as important regulators of liver regeneration. miR-21, miR-122, miR-126, and others modulate expression of pro- and anti-proliferative genes, fine-tuning the regenerative response. These miRNAs also serve as potential biomarkers for monitoring regeneration and predicting complications[62,63].

Chromatin remodeling proteins including high-mobility group box 2 (HMGB2) regulate hepatocyte proliferation by modulating chromatin accessibility. HMGB2 expression strongly correlates with cell cycle progression during regeneration, and its absence leads to delayed proliferation and compensatory hypertrophy[64,65].

Metabolic and functional recovery

Beyond cellular proliferation, functional recovery of hepatic metabolic capacity represents a critical aspect of regeneration. Hepatocyte hypertrophy, characterized by increased cell size without division, can compensate for parenchymal loss particularly in aged or injured livers where proliferative capacity is reduced. The balance between hyperplasia (increased cell number) and hypertrophy (increased cell size) varies based on age, underlying liver disease, and regenerative stimulus[66,67].

Markers of liver regeneration include alpha-fetoprotein, des-carboxy prothrombin, and ornithine decarboxylase, which increase during active regeneration. However, functional recovery measured by synthesis of albumin, coagulation factors, and metabolic capacity may lag behind volumetric restoration. This dissociation between volume and function underscores the importance of functional assessment in surgical planning[68,69].

Biomarkers predicting regenerative capacity remain an area of active investigation. Growth factors including HGF show promise as predictive markers, with preoperative HGF levels correlating with post-resection regeneration. Hyaluronic acid levels increase during regeneration and correlate with PHLF risk when elevated. Inflammatory cytokines IL-1 and IL-6 measured serially may provide early warning of inadequate regeneration[70,71].

Table 3 and Figure 4 summarize the molecular pathways and biomarkers in liver regeneration.

Figure 4 Sequence of events of liver regeneration after partial hepatectomy. TNF-α: Tumor necrosis factor-alpha; EGF: Epidermal growth factor; EGFR: Epidermal growth factor receptor; HGF: Hepatocyte growth factor; HSCs: Hepatic stellate cells; ECs: Endothelial cells; VEGF: Vascular endothelial growth factor; FGF1/2: Fibroblast growth factors 1 and 2; PDGF: Platelet-derived growth factor; BECs: Biliary epithelial cells; HB-EGF: Heparin-binding epidermal growth factor-like growth factor; ECM: Extracellular matrix; NF-κB: Nuclear factor kappa B; MMP-9: Matrix metalloproteinase-9.

Table 3 Molecular pathways and biomarkers in liver regeneration.

Pathway/factor	Main functional role	Clinical or experimental marker	Relevance in regeneration
TNF-α/IL-6	Initiation phase, hepatocyte priming	Serum TNF-α, IL-6	Predicts early regenerative activity
HGF/Met	Proliferation, cell signaling	Serum HGF	Predicts regenerative capacity
Wnt/β-catenin	Proliferation, stem cell activation	Nuclear β-catenin	Regulates proliferation vs differentiation
mTORC1	Controls hypertrophy vs hyperplasia	Phospho-S6K, mTOR gene	Promotes protein synthesis for regeneration
MicroRNAs (e.g., miR-21, miR-122)	Transcriptional modulation	Circulating miRNA signatures	Biomarker for regeneration or injury

TNF-α: Tumor necrosis factor-α; IL-6: Interleukin-6; HGF: Hepatocyte growth factor.

AI-DRIVEN OUTCOME PREDICTION AND RISK STRATIFICATION

PHLF prediction models

Accurate prediction of PHLF risk enables individualized surgical planning and patient counseling. Traditional scoring systems including Child-Pugh, model for end-stage liver disease, and albumin-bilirubin (ALBI) provide baseline assessments but lack the granularity to capture complex interactions between patient factors, tumor characteristics, and operative variables. ML models leveraging multiple data sources can improve predictive accuracy beyond conventional scores[72,73].

Multiple ML algorithms have been evaluated for PHLF prediction. A study by Tashiro et al[3] comparing 15 ML algorithms found extreme gradient boosting achieved the best performance with 93.1% accuracy and AUC of 0.863. This outperformed traditional ALBI (AUC 0.64) and FIB-4 (AUC 0.58) scores. Another multicenter study by Wang et al[33] using LightGBM algorithm achieved AUC values of 0.944 in training, 0.870 in validation, and 0.822 in testing cohorts.

Key predictive features identified by ML models include international normalized ratio, serum sodium, albumin, total bilirubin, platelet count, extent of resection, and presence of portal hypertension[74]. SHAP analysis provides interpretable explanations of individual predictions, identifying which factors most strongly influence PHLF risk for each patient. This explainability is critical for clinical acceptance and actionable decision-making[75,76].

Integration of imaging-derived features enhances prediction accuracy. A radiomics model based on gadoxetic acid-enhanced MRI combined with clinical variables [ALBI score, indocyanine green (ICG)-R15] achieved AUC of 0.84 for PHLF prediction. Radiomic features from the hepatobiliary phase reflect functional heterogeneity within the FLR, capturing information not apparent from visual assessment or volumetry alone[77,78].

Preoperative models enable risk stratification and surgical planning, while postoperative models incorporating early laboratory values provide dynamic risk assessment. Lactate and bilirubin on postoperative day 1 significantly enhance prediction, allowing early intervention for high-risk patients. Dynamic prediction models that update risk estimates as new data become available represent the next evolution in personalized postoperative care[79,80].

Radiomics and texture analysis

Radiomics extracts quantitative features from medical images that reflect tumor biology, tissue characteristics, and prognostic information invisible to visual inspection. In liver surgery, radiomics has been applied to predict treatment response, recurrence risk, and postoperative outcomes[81,82].

For CRLMs, radiomics features from preoperative CT predict histopathological characteristics including tumor growth pattern (expansive vs infiltrative), tumor budding, and resection margin status. A study by Granata et al[49] achieved 97% accuracy in identifying tumor growth pattern using seven texture features. This information guides surgical strategy, as infiltrative tumors require wider resection margins[83,84].

MRI-based radiomics, particularly using gadoxetic acid enhancement, provides additional functional information. EOB-MRI radiomics achieved 91% AUC for predicting recurrence after resection of CRLMs. Features extracted from the hepatobiliary phase reflect hepatocyte function and may identify occult micrometastases missed by conventional imaging (Figure 5)[85,86].

Figure 5 Radiomics analysis pipeline.

Radiomics also characterizes underlying liver parenchyma, detecting steatosis, fibrosis, and steatohepatitis that impact regenerative capacity. AI-assisted analysis of liver texture on CT or MRI correlates with histopathological steatosis grade, achieving AUC values of 0.91-0.95. Automated quantification of hepatic fat content using deep learning enables non-invasive risk assessment prior to major hepatectomy[87,88].

Integration of radiomics with clinical data in combined nomograms improves prediction accuracy for multiple outcomes including lymph node metastasis, recurrence, and survival. ML classifiers including random forest, support vector machines, and neural networks effectively integrate heterogeneous data types for personalized outcome prediction[89,90].

Table 4 summarizes the radiomic features which predict surgical outcomes.

Table 4 Radiomics features predicting surgical outcomes.

Outcome	Imaging modality	Key radiomics features	Best classifier	Performance (AUC/accuracy)
Recurrence (CRLM)	EOB-MRI	VIBE_FA10 textural metrics	KNN	AUC 0.91, Acc. 93%
Tumor front growth (exp. vs inf.)	CT	7 textural features (GLCM, etc.)	KNN	Acc. 97%, Sens. 90%, Spec. 100%
Tumor budding	Contrast MRI	11 textural features (arterial phase)	KNN	Acc. 95%, Sens. 84%, Spec. 99%
Early recurrence	CT	Tumor quality and quantity model	ML ensemble	AUC 0.83
Macrovesicular steatosis (donor)	CT	7 selected radiomic features	Logistic regression	AUC 0.87

CRLM: Colorectal liver metastases; EOB-MRI: Gadoxetic acid-enhanced magnetic resonance imaging; GLCM: Gray-level co-occurrence matrix; KNN: K-nearest neighbors; AUC: Area under the curve; CT: Computed tomography; ML: Machine learning; Acc: Accuracy; Sens: Sensitivity; Spec: Specificity; MRI: Magnetic resonance imaging.

Surgical decision support systems

Real-time intraoperative guidance represents an emerging application of AI in liver surgery. Computer vision algorithms can identify critical anatomical structures including bile ducts, hepatic vessels, and no-go zones from laparoscopic or robotic camera feeds. A proof-of-concept study by Madani et al[21] achieved mean accuracy of 0.80 for identifying critical structures and 0.95 for no-go zones during cholecystectomy.

AR systems overlay preoperative planning data onto the surgical field, compensating for organ deformation and providing real-time navigation[91,92]. These systems track surgical instruments and anatomical landmarks, updating the projected image to maintain registration despite tissue manipulation. Integration with fluorescence imaging using ICG enhances visualization of hepatic segments and tumors[14,93].

Decision support systems for surgical approach selection analyze patient characteristics, tumor features, and surgeon experience to recommend optimal strategies[94,95]. A web-based ML model developed by Wang et al[33] assists in choosing between anatomical and non-anatomical resection, achieving 0.82 accuracy for surgical decision prediction. The system additionally predicts overall survival and recurrence-free survival, enabling comprehensive preoperative counseling.

Future applications include real-time prediction of complications based on intraoperative data streams. Sensors monitoring hemodynamics, tissue oxygenation, and metabolic parameters could feed AI algorithms that alert surgeons to impending complications. Deep reinforcement learning may eventually enable adaptive surgical guidance that optimizes instrument trajectories and tissue handling based on real-time feedback.

EMERGING REGENERATIVE STRATEGIES AND THERAPEUTICS

Stem cell and growth factor therapies

Cellular therapies aimed at enhancing liver regeneration represent a promising frontier. Mesenchymal stem cells (MSCs) secrete paracrine factors including growth factors, cytokines, and extracellular vesicles that promote hepatocyte proliferation and reduce inflammation. A rodent study by Wu et al[14] demonstrated that hydrogel-encapsulated MSC-conditioned medium significantly accelerated liver regeneration following partial hepatectomy in steatotic livers.

MSC therapy addresses a critical challenge: Impaired regeneration in diseased livers including steatosis, fibrosis, and cirrhosis. In metabolic dysfunction-associated steatotic liver disease, baseline regenerative capacity is compromised, increasing PHLF risk[96,97]. MSC-CM treatment enhanced proliferating cell nuclear antigen expression, ATP production, and β-hydroxybutyrate content, indicating improved metabolic function[5,98].

HGF administration represents another therapeutic approach. Exogenous HGF enhances regeneration in preclinical models, though clinical translation has been limited by pharmacokinetic challenges. Gene therapy delivering HGF or other growth factors via viral vectors offers sustained local expression, potentially superior to bolus protein administration[99,100].

Hepatic progenitor cells, also termed oval cells, can differentiate into hepatocytes and cholangiocytes during severe injury when mature hepatocyte proliferation is impaired. Understanding factors that promote progenitor cell activation and differentiation may enable therapeutic strategies for patients with compromised hepatocyte proliferative capacity[101,102].

Pharmacological enhancement of regeneration

Small molecule enhancers of regeneration offer potential for simple clinical implementation. mTOR pathway activators can shift regeneration from proliferation-based to hypertrophy-based mechanisms, which may benefit elderly patients or those with impaired proliferative capacity. A study by Gielchinsky et al[103] demonstrated that mTOR activation mimicked the regenerative benefits of pregnancy in aged mice, dramatically improving survival after hepatectomy.

Nuclear receptor agonists including FXR agonists modulate bile acid signaling to enhance regeneration while reducing cholestasis[104,105]. PPAR agonists influence lipid metabolism and inflammatory responses, potentially improving regeneration in steatotic livers. Careful timing and dosing are critical, as these pathways also regulate termination of regeneration[103,106].

Anti-inflammatory approaches may reduce the excessive inflammatory response that contributes to PHLF. However, inflammation plays essential roles in initiating regeneration, so complete suppression could be counterproductive. Selective targeting of maladaptive inflammatory pathways while preserving beneficial responses represents a nuanced challenge[107,108].

miRNA therapeutics targeting specific miRNAs involved in regeneration are under investigation. Antagomirs (miRNA inhibitors) or miRNA mimics could modulate regenerative responses, though delivery to hepatocytes and off-target effects remain technical challenges[109,110].

SIRT and hypertrophy

Radioembolization with Yttrium-90 microspheres induces contralateral liver hypertrophy while providing tumor control, serving dual purposes in surgical candidates with borderline FLR. SIRT achieves median contralateral hypertrophy of 35%-47% within 2-3 months. Modified radiation lobectomy delivering higher doses (> 150 Gy) produces more pronounced hypertrophy while ablating tumors[111,112].

Compared to PVE, SIRT offers the advantage of tumor treatment during the hypertrophy induction period, potentially reducing dropout due to disease progression. However, radioembolization-induced liver disease (REILD) occurs in 2%-5% of cases when excessive radiation doses are delivered to non-target liver. Personalized dosimetry based on technetium-99 m macroaggregated albumin simulation reduces REILD risk while optimizing tumor dose[113,114].

Factors influencing hypertrophy after SIRT include baseline liver function, administered activity, volume of treated liver, and patient age. Younger patients with smaller spleen volumes, preserved liver function, and larger percentages of treated liver demonstrate greater hypertrophy. These parameters should be considered when selecting patients for SIRT as a bridge to resection[115,116].

Combining SIRT with PVE represents an emerging approach that may synergize hypertrophic effects. Sequential or simultaneous application of both modalities provides maximal deprivation of the deportalized liver while maintaining arterial perfusion to support tumor necrosis. Clinical data on this combined approach remain limited but preliminary results are promising[117,118].

CLINICAL INTEGRATION AND IMPLEMENTATION CHALLENGES

Workflow integration and validation

Successful clinical implementation of AI tools requires seamless integration into existing radiological and surgical workflows. Standalone AI applications that require separate data export, processing, and re-import create inefficiencies that limit adoption. Modern AI platforms integrate directly with picture archiving and communication systems and electronic health records, enabling automated analysis as part of routine imaging interpretation[119,120].

Clinical validation through prospective trials is essential before widespread adoption. Most published AI models are developed and validated retrospectively on single-institution datasets, which may limit generalizability. External validation in diverse populations and clinical settings is necessary to establish robustness. Multicenter prospective registries collecting standardized data facilitate rigorous evaluation of AI tools in real-world practice[121,122].

Regulatory approval pathways for AI medical devices are evolving. The FDA and European regulatory bodies have established frameworks for AI/ML-based software as medical devices. Class II devices requiring 510 (k) clearance or CE marking must demonstrate substantial equivalence to predicate devices or demonstrate safety and efficacy. Adaptive AI algorithms that learn from new data pose unique regulatory challenges requiring novel oversight mechanisms[123,124].

Ethical considerations and bias mitigation

Algorithmic bias represents a significant ethical concern in healthcare AI. Training data unrepresentative of diverse populations can produce models that perform poorly for underrepresented groups, potentially exacerbating health disparities. In liver surgery, most published AI models are developed on East Asian or European populations, with limited validation in African, Latin American, or other demographics[125,126].

Sources of bias include selection bias in patient cohorts, measurement bias in ground truth labels, and algorithmic bias in model architecture or training procedures. Mitigation strategies include diverse dataset curation, fairness-aware algorithms that explicitly optimize for equitable performance across groups, and algorithmic auditing to detect bias before deployment[127,128].

Transparency and interpretability are ethical imperatives for clinical AI. Black-box models that provide predictions without explanation undermine clinical trust and prevent identification of erroneous reasoning. Explainable AI techniques including SHAP, attention mechanisms, and layer-wise relevance propagation provide insights into model decision-making. These explanations enable clinicians to verify that predictions are based on medically relevant features rather than spurious correlations[38,129].

Data privacy and security require robust protections. Medical AI systems process sensitive patient data subject to regulations including GDPR in Europe and health insurance portability and accountability act in the United States. Federated learning approaches that train models on distributed datasets without centralizing data offer privacy-preserving alternatives for multi-institutional collaboration. Differential privacy techniques add controlled noise to protect individual patient information while maintaining model accuracy[19].

Cost-effectiveness and resource allocation

Economic evaluation of AI technologies is necessary to justify implementation costs. Initial investments include software licensing, hardware infrastructure, technical support, and training. Cost-effectiveness depends on downstream benefits including reduced complications, shorter hospital stays, and improved long-term outcomes[130,131].

AI-assisted surgical planning that changes operative approach in 44% of cases to optimize FLR preservation may reduce PHLF rates by 5%-10%, avoiding costs associated with intensive care, reoperations, and prolonged hospitalization. Predictive models that identify high-risk patients enable targeted interventions that may prevent complications. However, prospective economic analyses demonstrating return on investment are lacking for most applications[132,133].

Resource allocation considerations include whether AI tools should be universally available or concentrated at high-volume centers. Centralization enables expertise development and quality assurance but may limit access for patients in remote areas. Cloud-based AI services accessed remotely could democratize access, though internet connectivity and data transmission bandwidth may limit feasibility in resource-constrained settings[134,135].

Table 5 summarizes the ethical, regulatory and implementation challenges in AI.

Table 5 Ethical, regulatory, and implementation challenges.

Challenge area	Description	Example/impact
Algorithmic bias	Incomplete data, non-generalizable models	Less accurate for underrepresented groups
Transparency/explainability	Black-box predictions frustrate trust	Difficult to validate medical rationale
Generalizability	Overfit to single-center/population data	Poor results in new settings
Data privacy/regulation	Compliance with HIPAA/GDPR	Re-identification risks in cloud-based AI
Clinical validation	Lack of RCTs/prospective evidence	Hinders regulatory approval, uptake

HIPAA: Health insurance portability and accountability act; GDPR: General data protection regulation; AI: Artificial intelligence; RCT: Randomized controlled trial.

Study quality and risk of bias assessment

Since this review synthesizes heterogeneous evidence-including retrospective cohorts, single-center feasibility studies, technical validation papers, and preclinical models-a structured assessment of study quality was done (Table 6). Major domains evaluated included: (1) Study design; (2) Sample size; (3) Internal vs external validation; (4) Level of evidence; and (5) Risk-of-bias features (selection bias, overfitting, unblinded outcome assessment, protocol heterogeneity). AI segmentation and radiomics studies were generally rated as moderate-to-high risk of bias, mostly due to reliance on internal validation and curated datasets. Clinical PVE/LVD/ALPPS studies tended to be moderate evidence, limited by retrospective design and selection bias. Radiomics prognostic studies showed moderate risk, commonly lacking external validation and using small cohorts.

Table 6 Summary of study quality across major evidence categories.

Category	Typical study design	Sample size range	Key findings	Level of evidence	Risk of bias (summary)
AI segmentation/volumetry	Technical validation, retrospective imaging datasets	40-1200 scans	Dice 092-0.97 for liver segmentation	IV	High risk-internal validation only, curated datasets, protocol heterogeneity
ML prediction models for PHLF	Retrospective multicenter or single-center	300-25000 patients	AUC 0.82-0.94 for PHLF prediction	III-IV	Moderate risk -class imbalance, unblinded outcome measurement, overfitting risk
Comparative PVE/LVD/ALPPS studies	Retrospective cohorts, meta-analyses	60-1800	LVD > PVE hypertrophy; ALPPS fastest hypertrophy	II-III	Moderate risk -selection bias, inconsistent endpoints, non-standard hypertrophy intervals
Radiomics prognostic studies	Retrospective, mostly single-center	40-300	Predict recurrence, FLR dysfunction	IV	High-to-moderate risk -small samples, overfitting, rare external validation
Preclinical regenerative biology	Rodent and in vitro	n = 6-60 animals	Pathway-level mechanistic insights	V	Low-to-moderate risk-mechanistic but non-clinical

FLR: Future liver remnant; PVE: Portal vein embolization; LVD: Liver venous deprivation; ALPPS: Associating liver partition and portal vein ligation for staged hepatectomy; ML: Machine learning; AI: Artificial intelligence; AUC: Area under the curve; PHLF: Post-hepatectomy liver failure.

AI-GUIDED CLINICAL DECISION FRAMEWORK

AI-based predictions can refine selection of hypertrophy techniques by quantifying individualized PHLF risk and hypertrophy trajectories. A suggested framework is outlined below.

AI-supported decision algorithm

While these thresholds require prospective validation, explicit risk-based algorithms illustrate how AI outputs could concretely modify management-moving beyond descriptive prediction toward actionable decision support (Table 7).

Table 7 Artificial intelligence-supported decision algorithm.

Predicted parameter	Threshold	Suggested clinical action
Predicted PHLF risk	> 20%-25%	Avoid major hepatectomy; consider LVD or two-stage strategies
	10%-20%	Prefer LVD over PVE; optimize FLR function, nutrition
	< 10%	PVE adequate; proceed with resection after hypertrophy
Predicted hypertrophy insufficient after PVE (model forecast)	< 25% expected growth	Prefer primary LVD
Tumor progression risk during waiting period	High	Consider ALPPS or accelerated LVD
Functional FLR prediction (radiomics/MRI)	Below functional cutoffs	Avoid ALPPS; prefer staged or non-surgical options

LVD: Liver venous deprivation; ALPPS: Associating liver partition and portal vein ligation for staged hepatectomy; MRI: Magnetic resonance imaging; FLR: Future liver remnant; PVE: Portal vein embolization.

Comparative hypertrophy characteristics

Comparative hypertrophy data must be interpreted cautiously. PVE cohorts often include patients with borderline FLR and significant comorbidities, whereas ALPPS cohorts typically comprise highly selected cases with aggressive tumors or urgent need for hypertrophy. LVD datasets also vary in technique (simultaneous vs sequential) and embolic materials. These differences introduce substantial selection bias, making direct effectiveness comparisons challenging (Table 8).

Table 8 Comparative characteristics of portal vein embolization, liver venous deprivation, and associating liver partition and portal vein ligation for staged hepatectomy.

Feature	PVE	LVD	ALPPS
Typical study design	Retrospective cohorts	Retrospective + prospective pilot	Retrospective ALPPS registry
Sample size range	0-50	0-40	0-70
Median hypertrophy (%)	0-37	0-50	0-60
Time to hypertrophy (days)	2-21	8-10	0-7
Failure-to-resect rate (%)	0-5	0	0
90-day morbidity (%)	0-10	5-10	5-35
90-day mortality (%)	0	0	2-8
Key limitations	Collateral formation	More complex, higher dysfunction risk	Highest morbidity and selection bias

PVE: Portal vein embolization; LVD: Liver venous deprivation; ALPPS: Associating liver partition and portal vein ligation for staged hepatectomy.

FUTURE DIRECTIONS AND EMERGING TECHNOLOGIES

Multimodal data integration

Next-generation AI systems will integrate diverse data types including imaging, genomics, proteomics, and clinical variables to provide comprehensive patient characterization. Multi-omics approaches identifying molecular signatures predictive of regenerative capacity could enable precision patient selection for aggressive resection strategies[136,137].

Wearable sensors and continuous monitoring devices generate longitudinal data streams that could feed adaptive AI models updating risk predictions in real-time postoperatively. Integration of patient-reported outcomes and quality of life metrics would enable holistic assessment beyond traditional clinical endpoints[138].

Digital twins-computational models simulating individual patient physiology-represent an ambitious future direction. By incorporating patient-specific anatomy, liver function, tumor biology, and predicted responses to interventions, digital twins could enable virtual testing of surgical strategies before operating[41]. This would take personalized surgical planning to unprecedented levels[98].

Robotic surgery and AI synergy

Integration of AI with robotic surgical platforms offers potential for semi-autonomous or fully autonomous surgical tasks. Computer vision algorithms interpreting operative field anatomy could guide robotic instruments to optimal positions or execute standardized subtasks under surgeon supervision. Deep reinforcement learning could optimize instrument trajectories and tissue handling based on outcomes from thousands of prior procedures[101].

Haptic feedback systems enhanced by AI could detect tissue properties predictive of bleeding risk or tumor involvement, alerting surgeons to adjust technique. Force sensing at instrument tips combined with predictive models could prevent inadvertent injury to critical structures[107].

Standardization of surgical techniques through AI analysis of recorded procedures could identify best practices and reduce outcome variability. By analyzing thousands of surgical videos, computer vision algorithms can correlate specific surgical maneuvers with outcomes, generating evidence-based technical recommendations.

Precision regenerative medicine

Personalized regenerative strategies tailored to individual patient biology represent the ultimate goal. Pharmacogenomic profiling could identify patients likely to benefit from specific growth factor or small molecule therapies. Baseline assessment of regenerative capacity through serum biomarkers, hepatobiliary scintigraphy, and radiomics could stratify patients for adjunctive regenerative interventions[46].

CRISPR gene editing technologies may eventually enable correction of genetic defects impairing regeneration. Ex vivo gene therapy in explanted liver grafts could enhance their regenerative capacity before transplantation. While these approaches remain largely experimental, proof-of-concept studies demonstrate feasibility[50].

Tissue engineering and bioartificial liver devices could provide temporary hepatic support during the critical early postoperative period when PHLF risk is highest. Hepatocyte organoids or 3D-printed liver tissue might someday provide alternatives to regeneration for patients with inadequate regenerative capacity[44].

LIMITATIONS AND KNOWLEDGE GAPS

Technical limitations of current AI systems

Despite impressive performance in research settings, current AI systems have important limitations. Model generalizability remains a major concern, with many algorithms performing poorly on external datasets differing in scanner protocols, patient populations, or clinical practices. Variability in image acquisition parameters, contrast timing, and slice thickness can degrade segmentation accuracy.

Label noise and inconsistency in ground truth annotations limit supervised learning performance. Inter-observer variability in manual segmentation used to train models introduces uncertainty that propagates through the AI pipeline. Active learning approaches that iteratively refine annotations and self-supervised learning methods that don't require manual labels may address this limitation.

Computational resource requirements for training deep learning models remain substantial, potentially limiting development to well-funded institutions. Inference (applying trained models) is generally faster but still requires GPU hardware for real-time applications. Cloud-based AI services could democratize access but raise concerns about data privacy and internet connectivity dependence.

AI performance is strongly influenced by dataset heterogeneity. Variations in CT/MRI protocols, scanner types, slice thickness, gadoxetic acid timing, and population demographics can significantly alter model accuracy. Radiomics features are particularly sensitive to acquisition variability. Most current models lack external validation, which raises concerns that reported performance overestimates real-world utility. Overfitting and spectrum bias remain common, especially in small studies. Large, federated, multi-ethnic datasets with harmonized pipelines are necessary to ensure true generalizability of both segmentation and prediction models.

While several segmentation models report dice > 0.95 and ML models achieve AUC 0.82-0.94, these metrics largely reflect internal validation on single-center or curated datasets. Performance typically decreases in external or prospective validation due to scanner differences, contrast protocols, liver disease heterogeneity, and annotation variability. Thus, these results should not be interpreted as readiness for clinical deployment. Future studies must emphasize prospective, multi-institutional validation with harmonized imaging protocols and standardized ground truth definitions.

Clinical evidence gaps

Most AI applications in liver surgery remain at early stages of clinical validation. Level 1 evidence from randomized controlled trials is lacking for essentially all AI tools. Prospective studies demonstrating that AI implementation improves clinical outcomes beyond current standard of care are needed before widespread adoption.

Long-term outcome data are particularly sparse. While AI models predict short-term complications with reasonable accuracy, their performance for 5-year survival prediction is less well established. Recurrence patterns may differ from historical data used for training if surgical techniques or systemic therapies evolve, potentially degrading model performance over time.

Integration of AI into clinical decision-making pathways requires human factors research. How clinicians interact with AI predictions, when they appropriately override recommendations, and how to prevent automation bias are critical questions. Decision support systems must enhance rather than supplant clinical judgment.

Regenerative strategy optimization

Optimal timing, patient selection, and technique for portal flow modulation remain areas of active investigation. Standardized protocols for PVE, LVD, and ALPPS vary substantially across institutions. Comparative effectiveness research in prospective registries could identify best practices.

Predictive biomarkers reliably identifying patients who will fail to achieve adequate hypertrophy before undergoing PVE would enable triage to more aggressive approaches. Conversely, identifying rapid responders who could proceed to hepatectomy earlier would reduce the interval during which tumor progression may occur.

The molecular mechanisms determining hypertrophy magnitude and rate remain incompletely understood. Why some patients demonstrate robust regeneration while others with similar clinical characteristics fail to hypertrophy adequately is unclear. Single-cell transcriptomics and spatial proteomics may reveal heterogeneity in hepatocyte regenerative capacity that explains these differences.

CONCLUSION

AI is revolutionizing hepatic surgery through enhanced preoperative planning, risk stratification, and outcome prediction. Automated liver segmentation, 3D reconstruction, and virtual surgical planning enable precise volumetric assessment and surgical simulation. ML models predict PHLF with superior accuracy compared to traditional scoring systems, facilitating personalized surgical decision-making. Radiomics analysis extracts prognostic information from medical images that guides resection strategy and adjuvant therapy selection.

Portal flow modulation techniques including PVE, LVD, and ALPPS have expanded resection boundaries by inducing FLR hypertrophy. These approaches harness the liver's remarkable regenerative capacity, enabling curative resection in patients previously deemed inoperable. Understanding the molecular mechanisms of liver regeneration-including cytokine and growth factor signaling pathways-informs optimization of these techniques and development of adjunctive regenerative therapies.

The integration of AI with regenerative strategies represents a synergistic approach to expanding hepatic resection boundaries. AI-driven patient selection identifies ideal candidates for aggressive approaches while flagging high-risk patients requiring alternative strategies. Predictive modeling of hypertrophy response could enable personalized treatment algorithms selecting optimal portal flow modulation techniques. Real-time monitoring and adaptive management guided by AI may improve outcomes for high-risk patients.

Despite remarkable progress, significant challenges remain. Clinical validation of AI tools through prospective trials is essential before widespread adoption. Addressing algorithmic bias, ensuring transparency, and maintaining data privacy are ethical imperatives. Cost-effectiveness analyses must demonstrate value justifying implementation investments. Technical limitations including generalizability and computational requirements need resolution.

Future directions include multimodal data integration, digital twins for personalized surgical simulation, AI-robotic surgery synergy, and precision regenerative medicine tailored to individual patient biology. Continued multidisciplinary collaboration between surgeons, hepatologists, data scientists, and bioengineers will drive innovation in this rapidly evolving field. The ultimate goal is to safely maximize the number of patients who can undergo curative hepatic resection while minimizing complications and optimizing long-term outcomes.