Endoscopy International Open最新文献

Background and study aims: Reporting of colorectal polyp morphology using the Paris classification is often inaccurate. Multimodal large language models (M-LLMs) may support morphological assessment. This study aimed to evaluate the accuracy of an M-LLM (GPT-4o) in classifying colorectal polyp morphology compared with expert and non-expert endoscopists.

Patients and methods: We used the SUN dataset of colonoscopy videos from 100 unique colorectal polyps, each labeled with the validated Paris classification. An M-LLM (GPT-4o) classified five representative frames per lesion. Three expert and three non-expert endoscopists, blinded to one another, performed the same task. The primary outcome was accuracy in differentiating non-polypoid (IIa/IIc) from polypoid (Is/Ip/Isp) lesions. The secondary outcome was accuracy in differentiating sessile (Is) from pedunculated (Ip/Isp) lesions. Given the exploratory design, no multiplicity correction was applied; point estimates are presented with 95% confidence intervals (CIs), and P values are interpreted descriptively.

Results: M-LLM accuracy for differentiating non-polypoid from polypoid lesions was 73% (95% CI 63%-81%), comparable to experts (75%, 65%-83%; P = 0.84) and non-experts (77%, 68%-85%; P = 0.52), with similar sensitivity and specificity. Accuracy for differentiating sessile from pedunculated lesions was 55% (95% CI 42%-67%), lower than experts (76%; P = 0.02) and non-experts (77%; P = 0.01), primarily due to poor specificity (12% vs. experts 82% and non-experts 88%; P < 0.01 for both comparisons).

Conclusions: M-LLMs performed comparably to endoscopists in distinguishing non-polypoid from polypoid lesions but failed to reliably identify pedunculated morphology.

Background and study aims: The aim of this study was to validation the accuracy of major guidelines and their adopted unfavorable histologic factors for lymph node metastasis (LNM) in pT1 colorectal cancer (CRC). National Comprehensive Cancer Network (NCCN), Japanese Society for Cancer of the Colon and Rectum (JSCCR), European Society for Medical Oncology (ESMO), and French Intergroup Clinical Practice Guidelines (FICPG) were included.

Patients and methods: Retrospectively, 519 cases receiving oncological resection with lymphadenectomy were included. Factors and four pathological parameters--histological differentiation grades (HDG), lymphovascular invasion (LVI), depth of submucosal invasion (DSI)(>1000μm), and tumor budding (TB)--were analyzed for their impact on LNM incidence individually and in combination as defined by the guidelines.

Results: HDG, LVI, TB, and gender (female) are risk factors for LNM in multivariate analysis. All guidelines have significant ability to predict LNM ( P <0.001). NCCN and ESMO have similar performance, in terms of sensitivity (63.8%/68.1%) and specificity (62.7%/69.5%). JSCCR and FICPG had similar performance in terms of good sensitivity (100%/100%) and low specificity (25.6%/25.2%). The JSCCR/FICPG group had higher sensitivity and lower specificity and accuracy than the NCCN/ESMO group.

Conclusions: HDG, LVI, TB, and gender (female) are independent risk factors for LNM of T1 CRC. DSI is an excellent negative predictor, although not an independent risk factor. The NCCN/ESMO guideline has medium sensitivity and requires improvement. The JSCCR/FICPG guideline has perfect sensitivity but low specificity, thus exposing patients to many unnecessary surgeries. There is revision potential for current guideline factors, including those beyond current pathological ones, to improve LNM prediction accuracy.

Background and study aims: Artificial intelligence (AI) is expected to enhance the ability of endoscopists to detect gastric neoplastic lesions; however, its effectiveness among highly skilled Japanese expert endoscopists has not been validated. We developed a novel AI-assisted diagnostic tool for detection of gastric neoplastic lesions and evaluated its utility by comparing the diagnostic performance of endoscopists with and without AI assistance.

Patients and methods: Diagnostic performance of gastric neoplastic lesions without and with AI assistance was compared among 14 expert endoscopists and 12 non-expert endoscopists using an evaluation dataset consisting of 150 images containing neoplastic lesions and 350 images without lesions. A general linear mixed model was applied for comparative analysis. The primary outcome was to demonstrate superiority of sensitivity and non-inferiority of specificity among expert endoscopists using AI compared with those without AI. The significance level for sensitivity was set at 2.5% and the non-inferiority margin for specificity was defined as a log odds ratio of -0.25.

Results: Our AI demonstrated superiority in sensitivity (66.4% without AI vs. 83.5% with AI; odds ratio [OR] 2.562, 97.5% confidence interval [CI] 2.069-3.172) and non-inferiority in specificity (90.8% without AI vs. 92.9% with AI; OR 1.326, 95% CI 1.122-1.565) among expert endoscopists.

Conclusions: AI contributed to improved diagnostic performance even among Japanese expert endoscopists in detecting gastric neoplastic lesions. These findings suggest that the AI system may have potential to support consistently high diagnostic performance across varying levels of endoscopic expertise.

Background and study aims: Post-endoscopy esophageal adenocarcinomas (PEEC) challenge timely diagnosis of esophageal adenocarcinomas (OAs). This study aimed to determine prevalence of PEECs in Auckland region and elucidate the most plausible causes through a root-cause analysis framework.

Patients and methods: OA cases diagnosed in Auckland from 2013 to 2022 were retrieved from the New Zealand Cancer Registry (NZCR). Electronic clinical data were collected via the Regional Clinical Portal software. The primary outcome was PEEC prevalence, defined as OA diagnosed 6 to 36 months following an esophagogastroduodenoscopy (EGD) that failed to detect the cancer. Identified PEECs were classified into six categories.

Results: Among 633 OA cases, 45 (7.1%) were PEECs. A higher prevalence of PEEC was observed in patients with Barrett's esophagus (BE) (18.1% vs 2.7%), undergoing surveillance EGDs (52.6% vs 3.6%) and with early-stage cancers. Root-cause analysis delineated the PEEC causes, classified as follows: A (17.8%): lesion was identified, endoscopic assessment was adequate, follow-up was appropriately planned and executed, yet PEEC developed; B (17.8%): follow-up was delayed due to administrative factors; C (22.2%): follow-up decisions were inappropriate; D (22.2%): inadequate endoscopic assessment; E (11.1%): lesion was unidentified despite adequate assessment; and F (8.9%): lesion was unidentified and assessment was inadequate. Categories B, C, D, and F comprised 71.1% of cases deemed potentially avoidable.

Conclusions: Auckland's PEEC prevalence aligns with international post-endoscopy upper gastrointestinal cancer rates. Root-cause analysis underscores that a significant proportion of PEECs may be preventable with improved clinical practice.

Background and study aims: "Leaky gut," caused by increased mucosal permeability, plays a pivotal role in various diseases. However, few methods are available to evaluate intestinal mucosal permeability in the living human body. We established a novel method for evaluation of mucosal permeability using indigo carmine (IC).

Patients and methods: Subjects undergoing colonic endoscopy for screening of colon polyps or evaluation of ulcerative colitis (UC) severity were enrolled. IC was endoscopically sprayed in the cecum, and blood samples were obtained before spraying and at 30 and 60 minutes after. Serum IC level was analyzed by liquid chromatographer/mass spectrometer equipped with a Nexera HPLC system.

Results: In both the control (subjects screened for colon polyps) and UC groups, all subjects had their highest serum IC levels at 30 minutes after spraying. Serum IC level was significantly higher in UC patients than in the controls at both 30 and 60 minutes after spraying. In the UC group, serum IC levels at both 30 and 60 minutes were significantly higher in patients with a Mayo endoscopic subscore (MES) 1 at the cecum than in those with MES 0 in the same area.

Conclusions: Endoscopic spraying with IC is useful for evaluation of intestinal mucosal permeability.

Background and study aims: Although the endoscopic pressure study integrated system (EPSIS) is useful to evaluate competency of lower esophageal sphincter as a major part of the anti-reflux barrier, its availability is limited. This study aimed to assess whether gastric fold stretching during insufflation can predict intragastric pressure (IGP) without special equipment.

Patients and methods: A retrospective analysis included 33 patients who underwent esophagogastroduodenoscopy and EPSIS between June and July 2024. Gastric fold stretching along the greater curvature at the level of the cardia, observed in a retroflex view during insufflation, was compared with EPSIS results by reviewing recorded videos. Time ranges were defined as follows, and IGP was measured at the end of each range: Time range 1, until the ratio of longitudinal fold thickness to the groove width between folds reached 1:2; Time range 2, until the ratio reached 1:4; and Time range 3, until the folds or mucosal ridges were almost flattened. Variability was assessed using the coefficient of variation (CV), calculated as the standard deviation divided by the mean.

Results: Time ranges 1, 2, and 3 were fully observed in 100%, 97%, and 70% of patients, respectively. Mean IGPs at the end of Time ranges 1, 2, and 3 were 8.9, 11.1, and 17.7 mmHg, with CVs of 0.32, 0.28, and 0.08, respectively.

Conclusions: Flattening of gastric folds or mucosal ridges during insufflation is a reliable predictor of IGP. This finding may help identify patients with anti-reflux barrier dysfunction during regular endoscopic examination.