Search Results (110)

Background/Aims: While pathogenic germline variants play a critical role in pediatric cancer susceptibility, traditional clinical genetics primarily focuses on single-gene interpretations. Transitioning to a systems-level analysis of inherited variation can uncover shared biological vulnerabilities, informing genetic counseling, surveillance, and targeted therapeutics. This study aims to implement an unsupervised machine learning framework to identify and characterize Germline Mutation Signatures (GMS) across diverse pediatric malignancies, elucidating latent genomic patterns that reveal shared oncogenic mechanisms. Methods: We analyzed germline whole-exome and whole-genome sequencing (WES/WGS) data from a retrospective cohort of 420 pediatric cancer patients and matched non-cancer controls. Variants were deeply annotated to capture multi-dimensional features, including predicted pathogenicity, splice-site disruption, regulatory impact, population frequency, and sequence context. To enable robust modeling, we integrated an augmented feature set encompassing evolutionary constraint, loss-of-function intolerance, and compositionally normalized substitution spectra. These high-dimensional annotations were processed using a deep autoencoder for non-linear representation learning, followed by Gaussian Mixture Modeling (GMM) of the latent space. Results: The framework delineated 13 signatures (GMS1–GMS13), yielding an optimal Davies–Bouldin index of 1.051. These signatures map to fundamental biological processes, including DNA repair deficiencies, transcription-coupled damage, replication stress, and aberrant RNA regulation. Crucially, these GMSs transcend traditional tissue-of-origin classifications, manifesting across multiple distinct cancer types. This observation indicates convergent germline etiologies and suggests potential shared susceptibilities to pathway-directed therapies. Conclusions: The discovery of these cross-cancer signatures provides a scalable, biologically interpretable framework for decoding inherited pediatric cancer risk. While the therapeutic mapping networks identified are currently exploratory and serve as a hypothesis-generating foundation, this deep learning-driven paradigm establishes a robust basis for stratified precision medicine. Pending prospective clinical validation, this approach holds significant translational potential to move beyond single-gene paradigms toward unified, systems-level precision oncology strategies. Full article

Youth drug abuse is a persistent public health concern in China. Community-based drug rehabilitation (CBDR), the final three-year stage of China’s official rehabilitation system, aims to help people who use drugs (PWUD) reintegrate into society, but reintegration remains limited, particularly among young PWUD. This study explores the social reintegration challenges faced by young PWUD aged 18 to 35 during the CBDR stage in Guangzhou, China. Semi-structured interviews were conducted with 16 participants and analyzed using reflexive thematic analysis (RTA). Three themes were identified: stigma and relational struggles, socioeconomic marginalization and daily life disruption, and limitations of the CBDR service model. These challenges were not separate but reinforced one another, with difficulties in one domain spilling into others and narrowing the space in which reintegration could occur. These findings suggest that addressing the reinforcing linkages between stigma, economic hardship, and service limitations requires a more coordinated approach to CBDR service provision, one that integrates vocational support into relapse prevention, builds flexibility into surveillance procedures, and provides participants and their families with realistic, evidence-based information about the prospects of recovery. Full article

Oral squamous cell carcinoma (OSCC) frequently develops within chronically injured oral mucosa and may be preceded by clinically recognizable oral potentially malignant disorders (OPMDs), which provide an important window for cancer interception. This review examines how etiological exposures, persistent inflammation, and lesion-specific epithelial–stromal–immune interactions cooperate during the transition from mucosal injury to dysplasia, carcinoma in situ, and invasive OSCC. Major carcinogenic exposures, including tobacco, alcohol, and areca nut, are considered together with context-dependent contributors such as microbial dysbiosis, viral infection, and immune-mediated epithelial injury. At the molecular level, inflammation-driven oral carcinogenesis involves cytokine and chemokine amplification, oxidative and nitrosative stress, NF-κB and STAT3 activation, the COX-2/PGE₂ axis, genomic instability, field cancerization, epithelial–stromal crosstalk, angiogenesis, immune dysregulation, and epigenetic and non-coding RNA-mediated reprogramming. Emerging tools such as molecular risk assessment, liquid biopsy, optical imaging, spatially resolved profiling, and artificial intelligence-assisted models may improve identification of high-risk lesions, although most biomarkers require further prospective validation. Prevention should therefore integrate exposure control, biopsy-based diagnosis, local treatment when indicated, long-term surveillance, and trial-based precision strategies according to lesion risk, intervention window, and safety profile. This review supports a shift from lesion-centered management toward risk-adapted precision prevention in inflammation-driven oral carcinogenesis. Full article

Background/Objectives: Despite advances in the TNM staging system, prognostic heterogeneity persists in early-stage non-small cell lung cancer (NSCLC). Lymphatic invasion (LI) is a known marker of aggression, but its independent significance in the critical, low-risk Stage I, N0 subgroup—typically ineligible for adjuvant therapy—remains poorly defined. We hypothesized that LI acts as a powerful, yet hidden, risk factor in this highly favourable cohort. Methods: This retrospective cohort study included 988 consecutive patients who underwent curative anatomical resection for NSCLC. All patients underwent complete resection with pathologically confirmed negative surgical margins (R0 resection). Cases were staged according to the 9th Edition of the TNM Classification of Malignant Tumours (TNM-9) and grouped as LI-positive or LI-negative. A critical subgroup analysis focused on 347 truly low-risk patients (TNM Stage I, N0, no vascular or pleural invasion). Overall survival (OS) was evaluated using the Kaplan–Meier method and multivariable Cox proportional hazards models. Results: In the entire cohort (n = 988), LI was present in 40.9% of cases. LI positivity was an independent predictor of worse OS in multivariable analysis (HR: 1.520, 95% CI: 1.004–2.301, p = 0.048). In the low-risk subgroup (n = 347), the presence of LI resulted in a drastic survival divergence, with 5-year OS declining from 96.1% (LI-negative) to 83.8% (LI-positive). Multivariable analysis confirmed LI as an independent adverse prognostic factor in this subgroup (HR: 4.002, 95% CI: 1.567–10.221, p = 0.004). Conclusions: Lymphatic invasion is a robust, independent adverse prognostic factor in resected NSCLC. LI may identify a subset of early-stage N0 NSCLC patients who warrant closer postoperative surveillance and prospective evaluation for adjuvant treatment strategies. Validation in prospective cohorts is required before LI can be formally integrated into staging algorithms or treatment guidelines. Full article

Background/Objectives: Artificial intelligence (AI) is increasingly being applied across the infectious-disease pathway, from syndromic surveillance and imaging triage to etiologic support, antimicrobial stewardship, and prognostication. However, the maturity of evidence differs considerably across use cases, and apparent technical performance does not always translate into real-world clinical utility. Methods: This structured narrative review synthesizes current evidence on the principal clinical and public-health applications of AI in infectious diseases, with particular attention to external validation, workflow integration, economic implications, and governance. Results: The strongest near-term evidence supports narrow-AI applications linked to constrained workflows, especially tuberculosis chest-radiograph triage, selected host-response and antimicrobial-resistance prediction tools, and clinician-facing stewardship aids. By contrast, sepsis prediction illustrates how internal model performance may deteriorate on external validation and generate substantial alert burden when implemented in routine care. Economic evaluations are promising but remain predominantly model-based and context-dependent. Evidence for generative AI and large language models is still in an early phase, consisting largely of vignette studies, retrospective comparisons, and small single-center pilots rather than prospective outcome-based evaluations. Conclusions: Overall, the most realistic clinical role of AI in infectious diseases is augmentation rather than replacement: prioritizing scarce diagnostic capacity, shortening time to action, and improving antibiotic selection. Safe translation into practice requires, in order, external validation with local calibration, prospective impact assessment, and governance frameworks that address drift, accountability, transparency, and human oversight. Full article

Background: Artificial intelligence (AI) is increasingly being proposed for postharvest food-safety control of animal-source foods, but its practical value depends on whether models can support real decisions rather than only report high accuracy. Methods: This narrative review used a structured literature mapping of peer-reviewed work, mainly from 2020 to 2025, identified through database searches and citation tracking using combined terms for artificial intelligence, machine learning, animal-source foods, postharvest food safety, slaughterhouse inspection, cold-chain monitoring, traceability, authenticity, HACCP, validation, and regulatory applicability. Results: The most implementation-proximate applications are computer vision prescreening in slaughterhouses and processing plants, sensor- and IoT-based cold-chain surveillance, freshness and adulteration screening, and digital traceability systems. Across these areas, stronger evidence is associated with clearly defined control points, transparent reference methods, external or temporal validation, auditable data flows, and documented human oversight. The main weaknesses are single-site datasets, retrospective designs, incomplete reporting of reference methods, limited workflow testing, and insufficient attention to false alerts, fallback procedures, and governance. Conclusions: AI should be viewed as targeted decision support, not as a replacement for established food-safety control. Future studies should prioritize prospective, multi-site, workflow-embedded validation and show how alerts lead to documented corrective or verification actions. Full article

Pancreatic ductal adenocarcinoma (PDAC) remains a highly lethal malignancy. The identification and management of precursor lesions, particularly the increasingly common intraductal papillary mucinous neoplasms (IPMNs), pose a significant challenge, creating a profound clinical dilemma between intercepting pancreatic ductal adenocarcinoma and avoiding surgical overtreatment. This literature review aims to synthesize the latest evidence to facilitate a transition from purely morphology-based surveillance toward a biologically informed risk stratification paradigm. This approach could provide a personalized risk-stratification algorithm that optimizes therapeutic management and enables timely intervention for pancreatic cancer. By using PubMed, Embase, Scopus, and Web of Science, we analyzed and summarized key findings from recent literature (2020–2025), including cohort studies, mechanistic analyses, evidence-based guidelines, and systematic reviews on cyst fluid biomarkers (CEA panels, DNA/RNA sequencing), and emerging AI applications. Prospective and multicenter studies consistently report that NOD is independently associated with high-risk stigmata, cyst progression, and malignant transformation. Mechanistic research suggests a bidirectional interplay between the evolving neoplasia and pancreatic endocrine dysfunction. Updated guidelines underscore the need for more precise diagnostic algorithms. Recent work demonstrates that advanced cyst fluid markers—CEA panels, DNA/RNA sequencing, and multi-omic signatures—significantly improve diagnostic accuracy. Furthermore, explainable AI models show encouraging performance in predicting malignancy and assisting patient triage. Risk stratification in PCLs is shifting from morphology-based assessment toward integrated, multimodal approaches combining clinical, endocrine, imaging, molecular, and computational data. Recent evidence positions new-onset diabetes as a clinically accessible and biologically plausible marker of high-risk IPMNs. Similarly, molecular assays and AI-enhanced analytics provide an additional layer of diagnostic precision. The development of personalized risk prediction algorithms could improve early detection of malignancy while reducing unnecessary surgical resections. Full article

Background and Objectives: Despite the widespread adoption of HPV-based cervical cancer screening, the optimal triage strategy for women with low-grade cytological abnormalities and non-16/18 high-risk HPV (hrHPV) types remains debated. This study evaluated the impact of ASCCP risk-based triage strategies on colposcopy referral and biopsy outcomes in a large tertiary care center. Methods: This retrospective cross-sectional study included 2748 sexually active women aged 30–65 years who underwent colposcopy at Ankara Etlik City Hospital (January 2023–June 2025). Of these, 1932 met ASCCP criteria for cervical biopsy. Cytology results, HPV genotypes (16, 18, and other hrHPV types), and histopathological findings were analyzed. CIN3+ (CIN3, adenocarcinoma in situ, or invasive carcinoma) was the primary outcome. Multivariable logistic regression identified independent predictors, with model fit assessed by Nagelkerke R² and the Hosmer–Lemeshow test. Results: The mean age was 42.8 ± 8.1 years. The overall CIN3+ prevalence was 15.9% (308/1932). HSIL cytology was the strongest independent predictor of CIN3+ (adjusted OR 22.41, 95% CI: 11.28–44.52). HPV16/18 combined with HSIL or ASC-H cytology conferred the highest risk (adjusted OR 17.88–21.67). Women with ASC-US or LSIL cytology and non-16/18 hrHPV types had CIN3+ rates below 10%. Irregular screening history was also an independent predictor (adjusted OR 1.38). A risk-based triage approach suggested a potential reduction of approximately 29.7% in colposcopy utilization. However, this estimate applies exclusively to the biopsied subgroup and does not account for potentially undetected lesions in the 816 non-biopsied women enrolled in surveillance follow-up. Conclusions: HSIL cytology and HPV16/18 positivity represent the highest-risk profile for CIN3+ and should remain primary indications for colposcopy. Conversely, women with ASC-US or LSIL cytology and non-16/18 hrHPV types may be candidates for surveillance-based co-testing rather than immediate colposcopy, potentially enabling a resource-efficient reduction in unnecessary procedures within the biopsied cohort studied. Prospective validation in broader colposcopy-referred populations is needed before generalizing these findings to primary screening settings. Full article

Background: Breast cancer with isolated bone metastases at initial diagnosis represents a clinically distinct metastatic phenotype, often associated with more indolent biology and relatively favorable outcomes compared with visceral metastatic disease. The survival impact of resecting the intact primary breast tumor in de novo metastatic breast cancer remains controversial. In this study, we evaluated the association between primary breast surgery and overall survival (OS), defined as the time from diagnosis to death attributable to breast cancer, in patients presenting with isolated bone metastatic breast cancer. Methods: We performed a retrospective population-based cohort study using the Surveillance, Epidemiology, and End Results (SEER) database. Because the SEER variable for bone metastasis at diagnosis is available from 2010 onward and HER2-defined subtype information is available in the modern SEER era, the effective study period was defined as 2010–2021 rather than the full 2000–2021 SEER release period. Patients with breast cancer and isolated bone metastases at presentation, without evidence of lung, liver, brain, or other distant metastatic sites at diagnosis, were included. Demographic and clinicopathological variables, including age, sex, race, biologic subtype, histology, chemotherapy, radiotherapy, and primary breast surgery, were analyzed. Overall survival (OS), defined as the time from diagnosis to death attributable to breast cancer, was estimated using Kaplan–Meier methods and compared using the log-rank test. Independent prognostic factors were evaluated using multivariable Cox proportional hazards modeling. Results: A total of 6500 eligible patients were identified. Surgery of the primary breast tumor was performed in 1513 (23.3%) patients, and 62.8% received chemotherapy. Five-year overall survival (OS) was significantly higher among patients who underwent surgery than among those who did not undergo surgery (59.5% vs. 38.6%; p < 0.001). In multivariable analysis, primary breast surgery remained independently associated with improved OS (hazard ratio [HR] 0.54, 95% CI 0.48–0.62; p < 0.001). Age, histology, chemotherapy, radiotherapy, and biologic subtype were also associated with prognosis. Sex was not significant in the unadjusted analysis (p = 0.188), and the multivariable sex finding was interpreted cautiously because only 96 men were included. Conclusions: In this population-based cohort of patients with de novo breast cancer and isolated bone metastases, primary breast surgery was associated with improved survival among selected patients. However, this association should not be interpreted as causal, given the inherent limitations of observational registry data, including treatment selection, potential immortal-time bias, unmeasured metastatic burden, performance status, systemic therapy type and response, and local symptom burden, which are not fully captured in SEER. These findings support careful multidisciplinary consideration of local therapy in selected patients, while emphasizing the need for confirmation in prospectively designed studies. Full article

Background: Clinical artificial intelligence (AI) is rapidly moving from retrospective model development into prospective evaluation, implementation, and routine care. Existing reviews have addressed specific aspects of this transition, including monitoring, drift, implementation, governance, and human–AI interaction; however, these bodies of work remain methodologically and conceptually fragmented across different review traditions. Methods: We conducted a scoping review of review-level and review-oriented literature. We searched MEDLINE, Embase, Scopus, and Web of Science Core Collection from database inception to 28 February 2026. We charted review characteristics and conducted an inductive thematic synthesis of extracted review-level findings, while distinguishing operational, deployment-proximal, methodological, and conceptual/governance-oriented evidence. Results: We included 25 review-level publications spanning systematic, scoping, methodological, narrative, and governance-oriented reviews. Three major themes emerged. First, clinically important risks were consistently framed as socio-technical rather than purely algorithmic: trustworthiness depended not only on technical performance, but also on fairness, transparency, workflow fit, human oversight, and organisational readiness. Second, the included review literature consistently recommended post-deployment monitoring but showed limited operational maturity; monitoring methods, action thresholds, fairness surveillance, and corrective responses were weakly standardised, and mature evidence from activated systems in routine care remained sparse. Third, trustworthy implementation was increasingly framed as a lifecycle governance challenge extending beyond procurement and initial validation to include local validation, subgroup auditing, drift detection, controlled updating, incident response, and, where necessary, rollback or retirement. Discussion: The review literature suggests a persistent normative–operational gap, meaning that recommendations about what trustworthy clinical AI should require have advanced faster than evidence on how monitoring, updating, and governance are implemented in routine care. The strongest unresolved challenge is therefore not principal generation alone, but the translation of monitoring and governance expectations into actionable operational systems. Conclusions: Post-development trustworthiness in clinical AI should be understood as a lifecycle property, not a one-time technical achievement. Future work should prioritise stronger operational evidence, clearer reporting of deployment-proximal and post-deployment evaluation, methodological standardisation of monitoring metrics and thresholds, implementation research on feasible governance models, and evaluation frameworks for assessing post-deployment safety, fairness, accountability, and sustainability. Full article

Background: Immune dysregulation and clinical immunosuppression are biologically plausible contributors to thoracic aortic wall vulnerability through endothelial injury, protease-mediated extracellular matrix remodeling, vascular smooth muscle cell dysfunction, and impaired vascular repair. Yet, the clinical relevance of immunomodulated states to thoracic aortic aneurysm (TAA) incidence or growth and acute aortic syndromes remains undefined. Methods: This comprehensive review synthesizes clinical and translation evidence linking immunomodulated states in solid organ transplantation, autoimmune disease (predominantly systemic lupus erythematosus), HIV, and oncologic therapies to thoracic aortic dilation, aneurysmal progression, and acute aortic events. Principal Findings: Across transplant, autoimmune, and HIV cohorts, recurring themes include chronic immune dysregulation, endothelial dysfunction, proteolytic matrix remodeling, and impaired vascular repair capacity, although thoracic segment-specific longitudinal growth data remain limited and are often embedded within analyses of multiple vascular beds. In oncologic cohorts, aggregate analyses generally do not demonstrate uniform acceleration of aneurysm growth with malignancy or chemotherapy exposure, although agent-level models suggest that regimen-specific effects may be obscured in pooled estimates. Two studies most directly addressed our question in thoracic-relevant contexts reported (1) very low mean annual ascending aortic aneurysm growth (0.18 ± 0.64 mm/year) with no detectable association with chemotherapy or radiotherapy and (2) prior immunosuppressive/cytostatic chemotherapy exposure to be common in a proximal TAA surgical cohort (39.3%) without a clear difference in thoracic phenotype at presentation or postoperative outcomes. In HIV cohorts, available evidence supports modest but reproducible proximal aortic remodeling and a clinically meaningful aneurysm burden across vascular beds, yet definitive thoracic segment-specific natural history data remain limited. Conclusions: The available literature supports clinical vigilance and exposure-aware surveillance, while suggesting that thoracic aortic risk is unlikely to be uniform across immunosuppressive and cytotoxic therapies. Standardized, segment-specific longitudinal imaging with granular agent-level exposure characterization (dose, duration, sequencing, and combination regimens), consistent definitions of baseline diameter and growth, careful adjustment for key confounders, and prospective ascertainment of dissection/rupture and operative endpoints are needed to translate immunobiology into actionable risk stratification and long-term management strategies. Full article

Survivors of primary gastric lymphoma (PGL) face a significantly elevated and persistent risk of developing second primary malignancies (SPMs), with gastric adenocarcinoma representing the most frequent SPM and standardized incidence ratios reaching up to 16-fold above the general population. This excess risk persists for decades after initial treatment and is associated with increased cause-specific mortality compared to matched primary cancers. Among patients with PGL, approximately 5% develop gastric cancer (with two-thirds being metachronous), and nearly 15% harbor precancerous lesions including atrophic gastritis, intestinal metaplasia, and dysplasia. Beyond gastric malignancies, survivors also experience elevated rates of extra-gastric SPMs, particularly digestive system tumors (43%), respiratory cancers (21%), and urinary tract malignancies (13%). Key risk factors include treatment with immunochemotherapy or radiotherapy, advanced age, male sex, advanced stage at diagnosis, ulcerative-type lymphoma morphology, and persistent Helicobacter pylori (HP) infection. Patients receiving combined chemoradiotherapy demonstrate the highest SPM risk, particularly for gastric and pancreatic cancers. These findings underscore the critical importance of lifelong, risk-adapted surveillance strategies integrating both hematology and gastroenterology follow-up. Annual endoscopic surveillance is recommended for high-risk patients, with intervals adjusted according to lymphoma histology, HP status, and the presence of precancerous gastric lesions. Mandatory HP eradication with confirmation of response is essential for reducing gastric cancer risk. Future research priorities include prospective, standardized studies to better quantify SPM risk, validation of molecular and microbiological biomarkers for individualized risk stratification, and development of predictive models to enable personalized surveillance protocols and improve long-term outcomes in this vulnerable population. Full article

Background: Colorectal cancer (CRC) represents a major global health burden, accounting for roughly 10% of all newly diagnosed cancers and cancer-related deaths worldwide. According to the World Health Organization, it is the third most diagnosed malignancy and the second leading cause of cancer mortality. Postoperative complications remain a significant concern after CRC resection, occurring in up to 50% of patients and contributing to increased morbidity, mortality, prolonged hospitalization, and substantial healthcare expenditure. Artificial intelligence (AI) has emerged as a transformative tool in modern healthcare, offering advanced capabilities in predictive analytics, clinical decision support, and personalized perioperative management. Methods: This review systematically evaluates the application of AI, specifically machine learning (ML) and deep learning (DL) algorithms, in the prediction of anastomotic leak (AL) and other major postoperative complications. In this context, AI models are generally used to refine risk stratification and enhance surgical decision-making. Results: A total of 13 studies were included, encompassing 15,105 patients. Across these studies, ML and DL algorithms consistently outperformed conventional statistical models in forecasting postoperative outcomes. Conclussions: Current evidence suggests that AI has substantial potential to improve perioperative risk prediction, support intraoperative decision-making, and personalize postoperative surveillance in patients undergoing CRC surgery. Methodological limitations, including a high risk of bias, limited external validation, heterogeneous outcome definitions, and inconsistent reporting, necessitate more robust, prospective, multicenter research before widespread clinical adoption can be realized. Full article

Background/Objectives: Artificial intelligence (AI) is increasingly proposed to improve surgical planning, guidance, and postoperative surveillance. Yet many promising applications remain disconnected from the full surgical pathway and the feasible limitations of clinical deployment. In contrast to prior reviews that primarily catalog AI use cases, this review combines the literature to define the translational pathway—from label design through staged validation to workflow integration—required for clinically deployable computed tomography (CT)-based surgical AI. CT and particularly computed tomography angiography (CTA) are especially usable sources for surgical AI because they provide a standardized three-dimensional anatomic model that is already embedded in many clinical workflows. In autologous breast reconstruction, deep inferior epigastric perforator (DIEP) flap CTA offers an unusually strong model system: the anatomy is discrete, surgeon decisions are actionable, and downstream operative and postoperative outcomes are measurable. These characteristics make DIEP reconstruction suitable not only for technical model development, but also for exacting testing of how CT-based AI should be annotated, validated, displayed, and governed. Methods: This focused narrative review combines evidence across the surgical workflow, spanning preoperative planning and risk stratification, intraoperative support, and postoperative monitoring. Reporting standards, implementation frameworks, governance, and regulatory sources were also considered when directly relevant to clinical deployment. Results: Across the available literature on breast reconstruction with the DIEP flap, preoperative CTA has been associated with reductions in operative time of approximately 54–76 min in individual studies. Semi-automated perforator mapping can reduce review time from 2 to 3 h to approximately 30 min. Intraoperative extended-reality tools and surgeon-facing navigation systems illustrate the importance of the ‘last mile’ of translation, while postoperative monitoring models show how imaging-linked data can support a closed-loop learning system. Across these stages, recurring limits include target mismatch, weak external validation, protocol variability, inconsistent reporting, limited subgroup analysis, and inadequate integration of economic and governance considerations. Conclusions: We argue that the next important step is not a generic autonomous model, but a clinically deployable DIEP-CTA-AI program. The practical blueprint proposed here is staged: structured anatomical labels, separate imaging, surgeons’ decisions, and outcome reference standards, dense intermediate endpoints, retrospective and external validation, reader studies, prospective silent deployment, and workflow-impact assessment. If implemented in this way, DIEP flap CTA can serve as a practical blueprint for CT-based AI translation in surgery more broadly. Full article

Artificial intelligence (AI) is increasingly transforming gastrointestinal endoscopy by supporting lesion detection, lesion characterization, quality assessment, and clinical risk prediction. Hemostatic decision-making represents a particularly complex field for AI integration, as therapeutic decisions are often made rapidly in the presence of active bleeding, impaired visualization, unstable patients, and variable lesion accessibility. This review critically examines the current evidence for AI-assisted decision-making in gastrointestinal endoscopy and endoscopic hemostasis, with emphasis on gastrointestinal bleeding, prediction of hemostatic therapy requirements, bleeding-risk stratification, rebleeding prediction, transfusion support, and post-procedural monitoring. Available studies suggest that machine learning and deep learning models may outperform conventional scoring systems in selected retrospective or validation cohorts, improve recognition of high-risk lesions, support less experienced endoscopists, and contribute to more individualized management of non-variceal bleeding, variceal bleeding, and capsule endoscopy findings. However, prospective interventional evidence remains sparse, and most available models are limited by retrospective design, single-center datasets, incomplete external validation, black-box decision-making, heterogeneous reporting, workflow barriers, and uncertain cost-effectiveness. AI should therefore be regarded as an adjunctive decision-support tool rather than an autonomous replacement for clinical judgment. Its future value will depend on prospective multicenter validation, explainability, real-time usability, regulatory clarity, post-deployment surveillance, and evidence of improved patient-centered outcomes before widespread implementation in emergency endoscopy practice. Full article