Peritoneal metastasis is a common manifestation of advanced malignant tumors. Traditional imaging and cytology have insufficient sensitivity for detecting subcentimeter lesions, whereas Transformer models, leveraging self-attention mechanisms, exhibit unique advantages in global context modeling and multimodal fusion. These models have shown preliminary potential in the fields of imaging, pathological and molecular diagnosis, and prognosis prediction. However, existing studies are generally constrained by single-center, small-sample designs, limited cross-center generalizability, and a lack of prospective validation; model interpretability and the computational complexity of clinical deployment also constitute critical bottlenecks. In the future, adopting the pretraining-fine-tuning paradigm of large-scale medical foundation models, advancing pan-omics fusion and dynamic temporal modeling, and conducting prospective interventional clinical validation will be key pathways to propel Transformers from assistive tools toward core components of intelligent diagnosis and treatment for peritoneal metastasis. This paper summarizes and analyzes currently available research data, reviews the applications of Transformers in three major scenarios-imaging diagnosis, pathological and molecular subtyping, and prognosis prediction-focuses on analyzing translational bottlenecks such as data scarcity, model interpretability, and clinical deployment, and proposes corresponding solutions and future directions.
Objective: To investigate the clinical efficacy, safety, and post-resistance treatment strategies of ripretinib in the treatment of advanced gastrointestinal stromal tumors (GIST), and to provide evidence-based references for optimizing individualized treatment of advanced GIST. Methods: This is a single-center retrospective real-world study. Inclusion criteria: (1) GIST confirmed by histopathology; (2) disease progression after previous treatment with other targeted drugs including imatinib; (3) continuous administration of ripretinib for more than one month with at least one efficacy evaluation. Exclusion criteria: (1) Clinical data were seriously missing, and data such as patient dosage, efficacy evaluation results, and adverse reactions could not be obtained; (2) complicated with other serious underlying diseases, such as uncontrolled severe cardiovascular and cerebrovascular diseases, liver and kidney failure, etc.; (3) Allergic or severe drug intolerance to ripretinib. According to the above criteria, a total of 93 patients with advanced GIST treated in West China Hospital of Sichuan University from July 2021 to March 2025 were collected. In this study, evaluable patients were defined as those who met any of the following criteria: (1) baseline and follow-up imaging examinations were completed in our center with complete imaging data, and the long diameter of target lesions could be accurately measured and the efficacy could be evaluated; (2) The efficacy evaluation was completed in other hospitals, but the image data and report were complete, and the exact measurement data of target lesions could be obtained after review by the radiology department of our hospital. All patients with advanced GIST enrolled in the study initially received the standard dose of ripretinib (150 mg once daily). After disease progression (PD), the following treatment strategies could be selected after multi-disciplinary team (MDT) discussion: (1) the dose of ripretinib was increased from 150 mg once daily to 150 mg twice daily (ripretinib dose-escalation group); (2) The standard-dose ripretinib (150 mg, once daily) was combined with local treatment, including surgery, interventional therapy and radiotherapy (ripretinib combined with local treatment group); (3) standard-dose ripretinib combined with frontline TKI, including imatinib, sunitinib and regorafenib (ripretinib combined with frontline TKI group); (4) other TKI treatment, including imatinib, sunitinib, regorafenib and avapritinib (other TKI treatment group). The primary outcome was median progression-free survival (mPFS). Secondary outcomes included objective response rate [ORR, the proportion of patients with complete response (CR) or partial response (PR)], disease control rate [DCR, the proportion of patients with CR, PR, or stable disease (SD)], overall survival (OS), and adverse events. Results: A total of 93 patients with advanced GIST were enrolled. Among 59 evaluable patients, 6 had PR, 30 had SD, and 23 had PD, with ORR of 10.2% and DCR of 61.0%. With a median follow-up time of 13 (2-47) months, the median mPFS was 7.0 (95%CI: 4.9-9.1) months, and the 1-year OS rate was 86.1%. Genotyping analysis showed that the mPFS in the KIT exon 11 mutation group was significantly longer than that in the exon 9 mutation group (12.0 months vs. 4.0 months, P=0.007). No statistically significant difference in mPFS was observed among different recurrence/metastasis sites (liver, peritoneum, or liver + peritoneum, P = 0.149). A total of 60 patients experienced progression on standard-dose ripretinib. According to the subsequent treatment strategy, the patients were divided into 4 groups: ripretinib dose-escalation group (16 patients), ripretinib combined with local therapy group (9 patients), ripretinib combined with frontline TKI group (6 patients), and other TKI group (9 patients). The mPFS in these groups was 4.0 months (95% CI: 1.5-6.5), not reached, 4.0 (95%CI: could not be estimated) and 12.0 months (95%CI: 1.0-25.5), respectively. There was no significant difference in mPFS among the four groups (P=0.345). In terms of safety, no grade 3-4 serious adverse reactions were observed in the whole group, and dose-escalation or combination therapy did not increase toxicity. Conclusion: Ripretinib can serve as an effective treatment option for Chinese patients with advanced GIST after failure of first-line or multiple prior TKI therapies, and it demonstrates a favorable safety profile. Although no standard treatment is currently available for patients with resistance to ripretinib, dose escalation of ripretinib, combination with local therapy, combination with prior TKIs, or switching to alternative TKIs can be considered, and all regimens exhibit a favorable safety profile.
Objective: To construct a survival prognostic prediction model for whole-slide histopathological images (WSIs) of gastric cancer based on multi-instance learning with sequential feature reconstruction, and to explore its predictive value. Methods: Retrospectively collected hematoxylin-eosin stained WSIs and clinical prognostic data of gastric cancer patients treated at the 1st, 8th and 9th Medical Centers of Chinese PLA General Hospital from January 2015 to December 2022. The sample sizes of the three medical centers were 386, 284 and 216 cases, denoted as Dataset 1, Dataset 2 and Dataset 3 respectively. This study developed a multi-instance learning model based on sequential feature reconstruction, named sequence reordering and sparse autoencoder-based enhanced representation transformer for multiple instance learning (S²ERT-MIL). After extracting patch features of pathological images via a pre-trained model, the model reconstructed and optimized patch instance features through sequence reordering, regional feature enhancement, sparse autoencoding and cross-regional information fusion, and finally output patient-level survival risk scores. Five-fold cross-validation was adopted to evaluate model performance, and comparisons were conducted with representative MIL models including Attention-Based Multiple Instance Learning (ABMIL), Clustering-constrained Attention Multiple Instance Learning (CLAM), Dual-Scale Multiple Instance Learning (DSMIL), Dual-Tier Feature Distillation Multiple Instance Learning (DTFD-MIL) and Representation Refinement and Transformation Multiple Instance Learning (RRT-MIL). The concordance index (C-index) was used to assess the survival prediction performance of the model. Patients were divided into high-risk and low-risk groups according to the median risk score, and the Kaplan-Meier method with Log-rank test was applied to evaluate the risk stratification ability. Visualization analysis of high- and low-attention patches was also performed. Results: S²ERT-MIL achieved the highest C-index across all three datasets, with values of 0.7493±0.0486, 0.6758±0.0347 and 0.7232±0.0362 in Dataset 1, Dataset 2 and Dataset 3, respectively, all higher than those of the comparison models. Kaplan-Meier survival analysis showed that S²ERT-MIL stratified patients into high-risk and low-risk groups with significantly different survival outcomes in all three datasets. Attention-region visualization showed that high-attention patches were mainly derived from areas with dense tumor cells, complex tissue architecture or marked stromal reaction, whereas low-attention patches mostly represented low-information regions or areas relatively weakly associated with prognosis. Conclusion: S²ERT-MIL can predict patient survival prognosis based on H&E-stained gastric cancer WSIs and showed favorable risk discrimination and stratification performance across three datasets.
The gut microbiota, acknowledged as the human body's 'second genome', plays a pivotal role in maintaining health. Digestive tract reconstruction surgery profoundly alters the anatomical structure and physiological environment of the gastrointestinal tract, thereby inducing significant shifts in the intestinal microbiota. These microbial changes subsequently influence host physiological functions through metabolic, immune, neuroendocrine, and other pathways. For instance, Roux-en-Y gastric bypass surgery enriches short-chain fatty acid(SCFA)-producing Bacteroides, improving systemic insulin sensitivity. Conversely, pancreaticoduodenectomy leads to a marked enrichment of potential pathobionts such as Klebsiella and Clostridium, which may elevate the risk of infections and tumor recurrence. This review comprehensively summarizes the characteristic changes in the gut microbiota following various digestive tract reconstruction procedures and discusses their multifaceted impacts on host physiology, aiming to provide insights for future experimental research and clinical practice.
Objective: To analyze the implementation status of bariatric and metabolic surgery and the provincial resource distribution characteristics in Guangdong Province in 2025. Methods: Descriptive analysis was performed based on the aggregated data of bariatric and metabolic surgeries reported by 39 hospitals in Guangdong Province in 2025. Using the 2024 provincial data as annual control, we analyzed the total surgical volume, surgical procedure composition, distribution of performing hospitals, urban distribution, and application of novel procedures, with comparative analysis combined with external data (national, Shanghai, and international public databases). External reference materials included the 2024 Annual Report of China Obesity and Metabolic Surgery Database, the 2024 Annual Report of Greater China Bariatric and Metabolic Surgery Database, the Analysis of Quality Control Data of Bariatric and Metabolic Surgery in Shanghai from 2012 to 2023, the global registry report of the International Federation for the Surgery of Obesity and Metabolic Disorders (IFSO), and public data released by the American Society for Metabolic and Bariatric Surgery (ASMBS)/Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program (MBSAQIP). National and Shanghai data were used for horizontal regional comparison, while US data served as a reference for international industry trends. Results: A total of 1611 bariatric and metabolic surgeries were completed in Guangdong Province in 2025, representing an 18.3% increase compared with 1362 cases in 2024; the number of hospitals performing such surgeries rose from 22 to 39. Procedure composition: conventional laparoscopic sleeve gastrectomy (SG) accounted for 1,086 cases (67.4%); Roux-en-Y gastric bypass (RYGB) for 142 cases (8.8%); sleeve gastrectomy with transit bipartition (SG-TB) for 82 cases (5.1%); single-port SG for 87 cases (5.4%); one-anastomosis gastric bypass (OAGB) for 61 cases (3.8%); revisional bariatric surgery for 59 cases (3.7%); bypass stent for 73 cases (4.5%), and robot-assisted SG for 4 cases (0.2%). No cases of hydrogel balloon therapy or endoscopic sleeve gastroplasty (ESG) were reported. The top 10 hospitals ranked by surgical volume in Guangdong Province in 2025 showed an obvious hierarchical distribution. The cumulative volume proportions of the top 3, top 5 and top 10 hospitals reached 71.8%, 81.8% and 90.6%, respectively. In terms of urban distribution, Guangzhou, Shenzhen and Zhongshan contributed 95.8% (1543 cases) of the provincial total. The total number of bariatric surgeries nationwide in China reached 32 342 in 2024, while Shanghai recorded 1430 cases in 2023. Sleeve gastrectomy (SG) accounted for 75.6% of all national procedures in 2024 versus 87.9% in Shanghai in 2023. Roux-en-Y gastric bypass (RYGB) made up 4.7% of national surgeries in 2024 and 7.7% in Shanghai in 2023. There were 1234 one-anastomosis gastric bypass (OAGB) procedures nationwide in 2024, accounting for 3.8%. In the United States, 4587 ESG procedures and 1461 intragastric balloon placements were performed in 2023. Conclusion: The annual volume of bariatric and metabolic surgeries in Guangdong Province maintained an upward trend in 2025, with SG remaining the predominant procedure; however, the distribution of surgeries was highly concentrated by hospital and city. Compared with published national and Shanghai data, Guangdong Province recorded a relatively high surgical volume nationwide and featured a more diverse procedural spectrum. In contrast to international published data, the application volume of novel surgical techniques and new modalities performed in Guangdong Province remained limited at present.
Gastric cancer remains one of the most common malignancies in China, and early diagnosis and accurate staging are critical for improving patient prognosis. Conventional imaging-based diagnosis of gastric cancer is limited by physician experience, equipment variability, and subjective interpretation, resulting in insufficient detection of early lesions and inconsistent staging assessment. Artificial intelligence (AI), powered by deep learning, radiomics, and related technologies, has demonstrated substantial potential in lesion detection, staging assessment, treatment response evaluation, and prognostic prediction for gastric cancer. These advances may improve diagnostic performance and promote more homogeneous clinical practice. However, clinical translation remains challenged by data heterogeneity and data silos, limited model generalizability, insufficient interpretability, inadequate prospective clinical evidence and regulatory frameworks, and variable acceptance among clinicians. This article reviews the current applications and technical advances of AI in gastric cancer imaging, analyzes the key barriers to clinical implementation, and proposes practical strategies from four perspectives: data governance, technological innovation, clinical translation, and physician-AI collaboration. The aim is to provide a reference for promoting standardized, accessible, and clinically meaningful implementation of AI in precision diagnosis and treatment of gastric cancer.
Objective: To explore the feasibility, safety and oncologic value of para-aortic lymph node (PALN) dissection (including open, laparoscopic and robotic approaches) in colorectal cancer patients with clinically suspected PALN metastasis. Methods: A retrospective observational study was performed. Clinical data of 202 colorectal cancer patients who underwent synchronous PALN dissection at Fujian Medical University Union Hospital between 2011 and 2025 were retrospectively reviewed, including 40 open surgeries, 148 laparoscopic surgeries and 14 robotic surgeries. Perioperative surgical data were collected; postoperative complications were graded by the Clavien-Dindo classification. The metastatic pattern of PALN involvement were analyzed. Logistic regression analysis was performed to identify its risk factors. Survival analysis was conducted using the Kaplan-Meier method, and independent prognostic factors for overall survival (OS) were identified via Cox proportional hazards regression model. Results: Among all 202 patients, 106 (52.5%) had rectal tumors, 77 (38.1%) had sigmoid colon tumors, and 19 (9.4%) had descending colon tumors. All patients successfully completed PALN dissection. Postoperative complications occurred in 89 patients (44.1%), including 40 cases (19.8%) of chyle leak. According to the Clavien-Dindo classification: 4 cases (4.5%) were grade Ⅰ, 75 cases (84.3%) grade Ⅱ, 9 cases (10.1%) grade Ⅲ, and 1 case (1.1%) grade Ⅳ. No death occurred within 30 days after surgery. Postoperative pathology confirmed that the PALN positive rate was 33.7% (68/202). In the PALN-positive group, lymphovascular invasion was detected in 41 cases (60.3%, 41/68) and perineural invasion in 34 cases (50.0%, 34/68), which were significantly higher than those in the PALN-negative group [13.4% (18/134) and 25.4% (34/134), respectively], with statistically significant differences (all P<0.001). Multivariate Logistic regression analysis revealed that tumor N2 stage (OR=11.665, P<0.001), positive No.253 lymph nodes (OR=17.683, P<0.001), number of harvested regional lymph nodes (OR=0.946, P=0.002), number of resected PALNs (OR=1.106, P=0.002) and lymphovascular invasion (OR=4.522, P=0.007) were independent risk factors for PALN metastasis. A total of 170 patients who underwent simultaneous para-aortic lymph node (PALN) dissection between 2011 and 2022 were enrolled, with a median follow-up of 43 months. Among them, the 5-year OS was 41.1% in 54 patients with positive PALN, versus 86.8% in 116 patients with negative PALN, and the difference was statistically significant (P<0.001). The 5-year OS in the laparoscopic and robotic minimally invasive surgery group was 78.4%, which was significantly higher than that in the open surgery group (52.2%), with a statistically significant difference (P=0.003). Patients without postoperative complications had a 5-year OS of 77.1%, higher than those with Clavien-Dindo grade I-II complications (73.4%). The 5-year OS of patients with Clavien-Dindo grade I-IV complications was only 15.6%. Tumor location and preoperative treatment had no significant impact on OS (all P>0.05).Postoperative recurrence and metastasis developed in 48 patients (28.2%, 48/170), among whom 9 cases (5.3%, 9/170) had recurrent PALN metastasis. Multivariate Cox regression analysis demonstrated that open surgery (HR=2.099, P=0.044), more than 3 pathologically positive PALNs (HR=3.427, P=0.002), and postoperative Clavien-Dindo grade Ⅲ-Ⅳ complications (HR=6.436, P<0.001) were independent risk factors for poor OS after simultaneous PALN dissection. Conclusion: For colorectal cancer patients suspected of PALN metastasis in clinical practice, PALN dissection technique is safe and feasible, and perioperative complications are generally controllable. PALN cleaning can bring survival benefits to some patients with PALN metastasis, while attention should be paid to the prevention and control of serious postoperative complications to improve patient prognosis.
Artificial intelligence has rapidly advanced in recent years and is increasingly being incorporated into preoperative assessment, intraoperative assistance, and postoperative quality control in gastric cancer surgery. Radical gastrectomy is characterized by complex anatomical planes, frequent vascular variations, extensive lymph node dissection, and multiple intraoperative decision points. Artificial intelligence may provide surgeons with more intuitive, continuous, and quantifiable assistance through image recognition, multimodal prediction, three-dimensional navigation, and surgical video analysis. This review summarizes recent progress and potential clinical value of artificial intelligence in gastric cancer surgery from four aspects: anatomical recognition, metastasis risk assessment and intraoperative staging assistance, surgical navigation and decision support, and surgical quality control. Current studies suggest that artificial intelligence can help identify the pancreas, perigastric vessels, loose connective tissue, peritoneal metastases, surgical phases, and instrument movements. It may also contribute to risk stratification, preoperative planning, intraoperative warning, surgical training, and standardized quality assessment. However, clinical translation remains limited by insufficient high-quality annotated data, limited cross-center robustness, inadequate validation in real-world operative settings, and unresolved ethical and responsibility issues. Future multicenter prospective studies are needed to determine whether artificial intelligence can reliably improve surgical quality, oncological safety, and patient outcomes.
Artificial intelligence (AI) technologies are rapidly being applied across all segments of colorectal cancer management, including cancer screening, endoscopic diagnosis and treatment, imaging and pathological analysis, therapeutic decision-making, intraoperative assistance, and follow-up management. Nevertheless, substantial disparities exist among various AI systems in terms of task definition, data sources, validation methodologies, performance metrics, safety thresholds, and real-world clinical practicability. There is an urgent need to establish a dedicated evaluation framework tailored to clinical scenarios specific to colorectal cancer. To address this demand, the Colorectal Surgery Group of the Chinese Society of Surgery (Chinese Medical Association), the Colorectal Cancer Committee of China Anti-Cancer Association, and the Colorectal Surgeon Expert Group of the Surgeon Branch (Chinese Medical Doctor Association) jointly assembled a working group composed of specialists covering colorectal surgery, digestive endoscopy, medical imaging, pathology, medical oncology, radiation oncology, artificial intelligence, medical statistics, medical informatization, medical ethics, and regulations. Through systematic literature retrieval, collation of existing clinical guidelines and regulatory documents, expert letter consultations, and panel discussions, the working group formulated 13 recommendations graded by levels of evidence and strength of recommendation, and developed the Expert Consensus on Evaluation Standards for Artificial Intelligence Applications in Colorectal Cancer (2026 Edition). This consensus establishes a tiered evaluation framework centered on seven core dimensions: algorithmic performance, clinical safety, clinical efficacy, user experience, model interpretability, ethical compliance, and continuous supervision. It further clarifies targeted evaluation requirements for key AI application scenarios, including AI-assisted lesion detection during endoscopy, AI-aided diagnosis via imaging and pathological slides, generative AI and clinical decision support, intraoperative AI assistance, and real-world surveillance of AI systems.The consensus underscores that AI systems shall only serve as auxiliary clinical tools and must not replace clinicians to independently render diagnostic, therapeutic or surgical decisions. Prior to clinical deployment, all AI systems for colorectal cancer shall undergo rigorous validation commensurate with their risk classification, implement disease-specific safety red lines, mandate mandatory clinician review, and enforce full-lifecycle supervision throughout clinical use.
Colorectal cancer liver metastasis (CRLM) has high diagnostic and therapeutic complexity, and multidisciplinary team (MDT) has become its core diagnostic and therapeutic path. However, traditional MDT is faced with problems such as uneven distribution of expert resources and inefficient information integration. The rapid development of artificial intelligence (AI) technology provides a new path for the intellectualization of CRLM-MDT. This article systematically reviews the application status of AI in the whole process of CRLM-MDT, including AI-assisted disease assessment, prognostic risk stratification and efficacy prediction, clinical decision-making, intraoperative imaging and quality control, and automatic case follow-up. It also analyzes the current bottlenecks of AI application, such as insufficient multimodal integration, clinical transformation barriers, ambiguous decision-making boundaries and lack of evaluation system, and looks forward to the development direction of human-machine collaborative MDT 2.0. Studies have shown that AI can improve the accuracy and efficiency of CRLM diagnosis and treatment, but it is necessary to break through the reviews of technology and clinical integration to truly realize the transformation from technical feasibility to clinical applicability, and provide more high-quality personalized diagnosis and treatment services for CRLM patients.
We describe the technical architecture of the Huashan Multidisciplinary Intelligent Nexus for Decision Support in Digestive Malignancies (HS-MIND), developed using a reasoning-augmented large language model (ReAG-LLM) and a multi-agent system (MAS). We also summarize our approach to implementing the system at a single center and our experience using it in the perioperative management of older patients with digestive system malignancies. Built on a general-purpose medical AI platform, HS-MIND was customized for the management of digestive system malignancies and deployed on premises. Its core architecture comprises four layers: the data and knowledge layer, the reasoning augmentation and tool-use layer, the multi-agent collaboration layer, and the clinical application layer, with safety governance integrated throughout the workflow. By combining retrieval-augmented generation (RAG), controlled chain-of-thought (CoT) reasoning, tool-augmented reasoning, and multi-agent collaboration, the system grounds its outputs in evidence, integrates information across specialties, ranks alternative management pathways, and provides prompts for clinician review. In representative high-risk scenarios, HS-MIND can proactively identify perioperative risk factors, integrate information across specialties, and generate structured, evidence-grounded recommendations. HS-MIND has been used to synthesize clinical information and review risks before multidisciplinary discussions in 25 older patients with digestive system malignancies and multiple comorbidities. A module designed to detect conflicts involving cardiovascular risk or antithrombotic therapy was triggered in 5 cases. In 23 cases, the system's ranking of alternative treatment pathways was reviewed during multidisciplinary discussions and used to inform decision-making. Two cases were flagged for further review because of insufficient evidence or unresolved conflicts, with final adjudication by senior physicians. Using de-identified data from a previously treated 76-year-old patient with low rectal cancer and severe coronary artery disease, we illustrate the system's longitudinal reasoning while preserving the actual chronology of clinical events, thereby showing how it operates in a high-risk scenario involving competing clinical priorities. We also characterize the system's early clinical use across this single-center series of 25 cases, with the aim of informing clinical decision-making.
Esophagojejunostomy remains the most technically demanding step in laparoscopic total gastrectomy. This review systematically traces the spiral evolution of esophagojejunal anastomotic techniques from open circular stapling and laparoscopy-assisted circular stapling, to totally laparoscopic linear stapling (Overlap and π-shaped methods), and subsequently to the resurgence of circular stapling under totally laparoscopic and robotic-assisted platforms. Linear stapling achieves complete minimally invasive surgery by eliminating the auxiliary incision; however, it faces technical bottlenecks including limited applicability for high anastomosis, difficulty in common opening closure, and compromised anti-reflux mechanisms. Conversely, circular stapling has regained vitality on the totally laparoscopic platform through hand-sewn purse-string sutures, modified anvil insertion techniques, and robotic assistance, preserving the anatomical and physiological advantages of end-to-side anastomosis, particularly for advanced esophagogastric junction tumors. Based on domestic and international evidence together with our institutional experience, this article proposes an individualized selection framework from the dimensions of tumor location, patient body habitus, and surgeon expertise, and outlines future directions in instrument innovation, technical integration, and high-quality evidence generation, aiming to provide practical guidance for surgeons in selecting esophagojejunostomy approaches.
Artificial intelligence (AI) has witnessed rapid development in the field of digestive endoscopy, especially for colonoscopy, where its application is the most mature. Based on deep learning algorithms, three major technical systems have been established in clinical practice, including computer-aided detection (CADe), computer-aided diagnosis (CADx), and computer-aided quality assessment (CAQ). Numerous randomized controlled trials have verified its prominent advantages in improving adenoma detection rate, reducing missed diagnosis risks, and assisting pathological prediction. Nevertheless, such benefits have not been fully reproduced in real-world studies, highlighting practical challenges in translating AI efficacy "validated in laboratories" into "usefulness in clinical settings". Based on clinical practice of colonoscopy, this article systematically reviews the current applications of AI. From the perspectives of inherent technical limitations of AI models and subjective factors in human-machine collaboration, it analyzes the underlying causes limiting the clinical benefits of AI in real-world scenarios. Furthermore, this study proposes a dual strategy of "exploring scenario-driven differentiated deployment" and "optimizing human-machine collaboration". On one hand, the appropriate application scope of AI should be rationally defined, and differentiated deployment should be carried out according to clinicians' experience levels and clinical scenarios. On the other hand, systematic training is required to help clinicians form scientific understanding and reasonable expectations of AI and establish an efficient human-machine collaboration model, so as to facilitate in-depth integration of AI and endoscopic techniques and promote its wide clinical implementation.
Objective: To compare the incidence of gastroesophageal reflux disease (GERD) following sleeve gastrectomy (SG) alone versus SG with reestablishment of the acute angle of His (SG-His), and to evaluate the safety and clinical efficacy of the latter approach. Methods: A total of 124 obese patients admitted to Beijing Chaoyang Hospital, Capital Medical University from January 2023 to December 2023 were enrolled in this randomized controlled study and assigned to either the study group (SG-His, n=64) or the control group (SG, n=60). Surgical duration, postoperative complications and postoperative pain were recorded. Reflux symptoms were assessed using the Reflux Disease Questionnaire (RDQ) and Gastroesophageal Reflux Disease Questionnaire (GerdQ) scores preoperatively and at 1, 3, 6, and 12 months postoperatively. Weight loss outcomes were evaluated at 12 months post-surgery. Results: Twelve months Follow-up was completed by 56 patients in the SG-His group and 55 in the SG group, yielding a dropout rate of 10.5%. No significant differences were observed between groups in baseline characteristics, including reflux scores, body weight, BMI, and comorbidities (all P>0.05). Operative time, postoperative complication rates, and pain-related metrics were comparable between groups (all P>0.05). With regard to reflux outcomes in the per protocol set, the SG-His group exhibited significantly lower RDQ scores than the SG group at 1 month [1.0 (0.0-7.0) vs. 5.0 (1.0-9.0), P=0.015], 3 months [0.0 (0.0-4.0) vs. 6.0 (0.5-9.0), P<0.001], 6 months [0.0 (0.0-2.0) vs. 4.0 (0.0-9.0), P<0.001], and 12 months [0.0 (0.0-4.0) vs. 2.0 (0.0-7.0), P=0.020] postoperatively. In the per protocol set, the GerdQ score was also significantly lower in the SG-His group at 6 months [6.0 (6.0-6.8) vs. 6.0 (6.0-8.0), P=0.014], although no significant differences were found at 1, 3, or 12 months (all P>0.05). According to RDQ criteria, the proportion of patients diagnosed with GERD was significantly lower in the SG-His group at 1 month [8.2% (5/61) vs. 21.1% (12/57), P=0.047], 3 months [3.5% (2/57) vs. 19.6% (11/56), P=0.007], 6 months [1.8% (1/56) vs. 17.9% (10/56), P=0.004], and 12 months [1.8% (1/56) vs. 14.5% (8/55), P=0.034]. Similarly, based on GerdQ criteria, GERD diagnosis rates were significantly reduced in the SG-His group at 3 months [21.1% (12/57) vs. 41.1% (23/56), P=0.021], 6 months [10.7% (6/56) vs. 30.4% (17/56), P=0.010], and 12 months [14.3% (8/56) vs. 36.4% (20/55), P=0.007]. There were no significant differences in weight loss between the two groups at 12 months post-surgery (P>0.05). Conclusions: Sleeve gastrectomy with reestablishment of the acute angle of His significantly reduces the incidence and severity of postoperative gastroesophageal reflux, effectively mitigates GERD symptoms, and maintains equivalent weight loss outcomes compared to SG alone. This technique is a safe and effective surgical technique.
Relying on core technologies including deep learning, radiomics and large language models, artificial intelligence (AI) has covered the full workflow of diagnosis and treatment in gastrointestinal surgery. Meanwhile, continuous iteration and large-scale clinical adoption of surgical robots are driving gastrointestinal surgical procedures toward higher precision and minimal invasiveness. The deep integration of artificial intelligence and surgical robots has reshaped the paradigm of gastrointestinal surgical care, facilitating the disciplinary transformation from conventional minimally invasive surgery to digital, intelligent and precise surgery. This paper reviews cutting-edge advances in robot-assisted gastrointestinal surgery empowered by artificial intelligence, analyzes prevailing bottlenecks hindering clinical translation, including insufficient data standardization, algorithm trustworthiness, and incomplete supporting regulatory and ethical frameworks. Combined with the latest domestic and international research outcomes, it prospects future development pathways, providing references for formulating standards and standardized clinical implementation of intelligent technologies in gastrointestinal surgery. The profound combination of AI and surgical robots breaks through the technical limitations of traditional minimally invasive surgery, bringing revolutionary transformations to the advancement of precise and intelligent gastrointestinal surgery. Gastrointestinal surgeons will not be directly replaced by AI in the future; instead, they will be superseded by peers proficient in AI and robotic technologies, and the evolution of gastrointestinal surgery will inevitably be led by AI-competent clinicians. With joint efforts across multiple disciplines, robot-assisted surgery in the AI era will undoubtedly elevate gastrointestinal surgery to an advanced new level.
Objective: To explore the clinical application of totally laparoscopic lymph node dissection around the lower esophagus and in the inferior mediastinum combined with digestive tract reconstruction via the esophageal hiatus and right thoracic approach for adenocarcinoma of the esophagogastric junction (AEG). Methods: A retrospective analysis was performed on 4 patients with locally advanced Siewert type Ⅱ/Ⅲ AEG without distant metastasis. A combined exposure technique, including traction of the left diaphragmatic crus using barbed sutures and deliberate opening of the right pleura, was applied under laparoscopy to create an adequate operative space for lymph node dissection around the lower esophagus, inferior mediastinal lymphadenectomy and digestive tract reconstruction. Results: A total of 4 patients underwent this surgical procedure from November 2024 to September 2025, including 3 cases of Siewert type Ⅱ and 1 case of Siewert type Ⅲ. There were 3 patients with clinical T3 stage and 1 patient with clinical T4b stage, and the tumor diameter ranged from 1 cm to 10 cm. Two patients received neoadjuvant therapy preoperatively. All operations were completed totally laparoscopically without conversion to open laparotomy or thoracotomy. The operative time ranged from 243 to 340 minutes, and intraoperative blood loss was 50 to 100 ml. The number of harvested lymph nodes was 17 to 30. One patient developed right pleural effusion (Clavien-Dindo grade Ⅲa) postoperatively, which resolved after percutaneous puncture and drainage without thoracic infection. No other complications occurred in the remaining patients. Conclusion: The totally laparoscopic approach via the esophageal hiatus and right thorax serves as a novel alternative for minimally invasive radical resection of high-position AEG.
Objective: To evaluate the efficacy of avapritinib as neoadjuvant therapy in gastrointestinal stromal tumor (GIST) with platelet-derived growth factor receptor α (PDGFRA)- D842V mutation. Methods: We conducted a retrospective analysis of patients with gastrointestinal stromal tumors harboring the PDGFRA-D842V mutation who received neoadjuvant treatment with avapritinib. Clinical and pathological data were retrieved from 7 medical institutions [Union Hospital of Tongji Medical College, Huazhong University of Science and Technology (3 cases); Fudan University Shanghai Cancer Center (3 cases); Peking University Cancer Hospital & Institute (2 cases); Peking University People's Hospital (1 case); Sun Yat-sen University Cancer Center (1 case); the First Affiliated Hospital of Chongqing Medical University (1 case); the Affiliated Hospital of Qingdao University (1 case)] from October 2019 to March 2025, including the patients' demographic data, laboratory and imaging findings, neoadjuvant therapy history, operative details, pathologic and genetic testing results, as well as postoperative adjuvant treatment and follow-up information. Results: After excluding 1 patient with concomitant gastric cancer, 7 patients presenting with distant metastasis at initial diagnosis or a prior history of GIST surgery, and 2 patients with incomplete clinical data, a total of 12 patients with PDGFRA-D842V mutations who received avapritinib neoadjuvant therapy were enrolled in this study. There were 8 male patients and 4 female patients, with a mean age at diagnosis of (56.0±12.4) years. Among them, 5 cases received a dosage of 300 mg/day, 2 cases received 200 mg/day, 2 cases received 150 mg/day, and 3 cases received 100 mg/day. The maximum tumor diameter before and after neoadjuvant therapy was (15.7±4.9) cm vs. (10.1±4.1) cm, respectively (t=5.998, P < 0.001), and the median time to best response was 4.5 (3.3,5.0) months. Among the 12 patients, 9 achieved partial response and 3 achieved stable disease, yielding an overall disease control rate of 100%. Common adverse events included anemia (8 cases), elevated bilirubin (6 cases), periorbital edema (5 cases), leukopenia (4 cases), and decreased neutrophil count (4 cases). Grade ≥3 adverse events occurred in eight patients, primarily anemia (2 cases), elevated bilirubin (2 cases), and leukopenia (1 case). All these adverse events resolved following appropriate supportive management. After achieving maximal response to neoadjuvant therapy, all 12 patients underwent surgical resection with curative intent. The median operative time was 95.0 (82.5, 120.0) minutes, and the median intraoperative blood loss was 50.0 (20.0, 175.0) ml. Among them, 7 patients underwent laparoscopic surgery, and 5 underwent open surgery. A total of 2 patients in the entire cohort underwent combined organ resection, including partial hepatectomy and partial transverse colectomy. The median postoperative length of hospital stay was 8.0 (7.0, 9.8) days, and no Clavien-Dindo grade III or higher complications occurred. Postoperatively, 5 patients continued treatment with avapritinib. The median follow-up duration was 20.5 (15.3, 37.3) months. The 2-year disease-free survival and overall survival rates were 68.2% and 77.9%, respectively. Conclusions: Avapritinib as neoadjuvant therapy in patients with PDGFRA-D842V-mutant gastrointestinal stromal tumors demonstrates significant efficacy in reducing tumor size and improving the rate of complete resection. Despite a relatively high incidence of adverse events, the treatment exhibits a manageable safety profile and is generally well tolerated.
Anastomotic leakage and anastomotic stenosis are among the most severe complications of rectal resection, among patients receiving neoadjuvant radiotherapy, the incidence of anastomotic stricture ranges from 3% to 6%. Anastomotic stricture can lead to abdominal pain and bowel obstruction, significantly affecting the quality of life and sometimes necessitating the creation of permanent stomas. This article reviews the definition of rectal anastomotic stricture. The development of anastomotic stricture is closely related to multiple factors, such as the healing characteristics of the bowel wall, local ischemia, anastomotic leakage, and the effects of preoperative radiotherapy. Furthermore, the choice of surgical anastomosis techniques significantly impacts the incidence of strictures. To enhance treatment outcomes for anastomotic strictures, this review summarizes different therapeutic strategies for membranous and tubular strictures. Transanal endoscopic resection and reconstruction surgery has significant advantages in the treatment of severe rectal anastomotic stenosis, providing standardized suggestions and guidance for the treatment of this clinical difficulty.
Objective: To investigate the feasibility and efficacy of ex vivo liver resection and autologous liver transplantation (ERAT) in the treatment of initially unresectable colorectal cancer liver metastases (CRLM). Methods: Four databases, including PubMed, Embase, Scopus and Web of Science, were systematically searched for literature on ERAT for CRLM published from database inception to May 13, 2025. Meanwhile, clinical data of CRLM patients who underwent ERAT at the Department of Hepatopancreatobiliary Surgery, The Second Affiliated Hospital of Zhejiang University School of Medicine between 2021 and 2024 were collected. Perioperative conditions and prognosis of the patients were summarized. Results: (1) Literature review: A total of 13 patients with CRLM who underwent ERAT from 5 studies were included. Among the 9 patients with available margin status, 8 achieved R0 resection and 1 underwent R1 resection. Among cases with available complication data, postoperative complications occurred in 7 patients, including pleural effusion in 3 patients. Other complications included infectious bile leakage, small bowel perforation, and inferior vena cava obstruction or compression. One patient died on postoperative day 15 due to acute respiratory failure and renal failure secondary to hemopneumothorax. Overall survival ranged from 15 days to 76 months. Among patients with available follow-up data, tumor recurrence occurred in 3 patients, including 1 patient with confirmed bone metastasis, while the sites of recurrence in the remaining patients were not clearly reported. (2) Institutional experience: All 5 patients in our center achieved R0 resection. One patient died of acute liver failure postoperatively, and the remaining cases recovered well. Pleural effusion occurred in all patients postoperatively, but the length of hospital stay did not exceed 3 weeks in any case. One patient had early postoperative recurrence and died at 27 months, while the other 3 patients remained alive at the last follow-up(2025/9/30). Conclusions: ERAT may serve as a potential surgical option for selected patients with unresectable CRLM, and radical resection can be expected under strict patient selection criteria.
The treatment of locally advanced (LARC) and locally recurrent (LRRC) rectal cancers poses significant challenges due to the anatomical complexity and the aggressive nature of tumor invasion. The adoption of multidisciplinary team (MDT) treatment models has become key to improving patient outcomes. Within the MDT framework, advances in imaging and pathology facilitate accurate assessment of disease. In minimally invasive surgery, urinary system reconstruction and pelvic floor reconstruction techniques have significantly improved outcomes for patients undergoing surgery. MDT decision-making plays a particularly important role in the selection of neoadjuvant treatment strategies: The MDT must weigh up the benefits and risks while taking into account the patient's primary disease, as well as their physical and mental condition, and strictly adhere to the indications for pelvic exenteration (PE) surgery. For patients without mesorectal involvement, neoadjuvant chemotherapy alone has demonstrated comparable efficacy to neoadjuvant chemoradiotherapy while exhibiting lower toxicity, but its application in T4b patients requires further validation. In conclusion, the treatment of LARC/LRRC has entered the era of multidisciplinary precision, and the MDT model is the core mechanism for integrating technological innovation and evidence to continuously improve patients' survival and quality of life. The future direction of development under the MDT model focuses on the integration of imaging and liquid biopsy for precise stratification, the optimization of the cost of robotic surgery and the innovation of bioprosthetic materials, the clarification of the optimal preoperative plan through multicenter studies, and exploring immune-based/targeted combination strategies.
