Digestive endoscopy : official journal of the Japan Gastroenterological Endoscopy Society最新文献

Background and aim: Endoscopic prediction of colorectal cancer (CRC) invasion depth is essential for determining optimal treatment strategy. Artificial intelligence (AI) may assist in distinguishing between superficial (Tis/T1a) and deeply invasive (T1b) lesions to avoid unnecessary surgery. In this narrative review, we summarize recent advances in this field.

Methods: A database search of PubMed, Scopus, Cochrane Library, and ClinicalTrials.gov was conducted in September 2025 to identify original peer-reviewed studies that developed or validated AI-based models using endoscopic imaging for invasion-depth prediction. Studies reporting diagnostic metrics such as sensitivity, specificity, accuracy, and AUC were included.

Results: Ten studies met the inclusion criteria, categorized into three groups: (1) AI prediction using image-enhanced endoscopy, (2) AI prediction using white-light imaging, and (3) AI prediction using multi-modal data (imaging with clinical information). The latest models achieved high performance for T1b CRC diagnosis (e.g., AUC 0.851), with some demonstrating performance comparable to expert endoscopists. However, a pooled analysis was not performed due to dataset heterogeneity and limited sample sizes.

Conclusions: AI-assisted systems show promise for improving the prediction of invasion depth in CRC and supporting real-time decision-making. However, limited sample sizes for training and test datasets and an imbalance in the training dataset remain key challenges. Large-scale, multicenter validation studies and the development of open-access databases are essential for clinical implementation.

Multimodal large language models (MLLMs) can automatically analyze clinical video, but evidence from full esophagogastroduodenoscopy (EGD) and the impact of on-screen computer-aided detection/diagnosis (CAD) overlays on MLLM behavior remain unclear. We tested whether an MLLM can produce clinically adequate EGD reports and whether a CAD overlay changes performance. We analyzed five complete EGD videos with Gemini 2.5 Pro in paired versions: (1) clean video and (2) the same video with a CAD overlay. Five blinded endoscopists rated report adequacy in three domains. MLLM accuracy for landmarks/lesions was further assessed by two blinded expert endoscopists using the time-window rule (a model detection counted as correct if it occurred within ±2 s of the expert-annotated timestamp). In this retrospective pilot study, five archived diagnostic EGD procedures from five patients were available as full-length videos. Across five raters, MLLM Completeness was judged adequate in 56.0% (14/25 ratings) with Clean-Video versus 48.0% (12/25 ratings) with Overlay-Video (p = 0.500). Visualization was identical (36.0% [9/25 ratings] for both; p = 1.000). Lesions characteristics were identical (16.0% [4/25] for both; p = 1.00). For the Landmark agreement, the overall accuracy of the MLLM with Clean-Video vs. Overlay-Video was: 0.55 [95% CI 0.43-0.67] vs. 0.33 [0.23-0.46], p = 0.029; sensitivity 0.53 [0.40-0.66] vs. 0.35 [0.24-0.49], p = 0.122; specificity 0.67 [0.35-0.88] vs. 0.22 [0.06-0.55], p = 0.125. In this pilot study, Gemini 2.5 Pro demonstrated inadequate performance for clinical EGD reporting. These hypothesis-generating findings suggest substantial optimization and larger-scale validation are required before deployment.

Objectives: Enteroscopy-assisted endoscopic retrograde pancreatography-guided pancreatic duct drainage (eERP-PDD) and endoscopic ultrasound-guided pancreatic duct drainage (EUS-PDD) are minimally invasive alternatives to surgery for pancreaticojejunostomy stricture (PJS); however, comparative data remain limited. We compared the effectiveness and safety of these approaches and identified factors associated with technical failure.

Methods: This multicenter retrospective study included 88 patients (111 procedures) who underwent endoscopic intervention for PJS at 13 Japanese tertiary centers. We compared clinical outcomes between eERP-PDD and EUS-PDD. The primary outcome was technical success; secondary outcomes included clinical success, procedure time, and adverse events (AEs). Propensity-score overlap weighting was used to adjust for baseline differences.

Results: As initial treatment, 77 patients underwent eERP-PDD and 11 underwent EUS-PDD. After adjustment, EUS-PDD achieved higher technical success (eERP-PDD, 28% vs. EUS-PDD, 71%; p = 0.012) and clinical success (22% vs. 71%; p = 0.003), with shorter procedure time (76 min vs. 41 min; p = 0.001). AE incidence was higher with EUS-PDD before adjustment (5% vs. 27%; p = 0.039) but comparable after adjustment (7% vs. 29%; p = 0.15); all AEs resolved with conservative management. Age < 75 years, male sex, and main pancreatic duct (MPD) diameter ≥ 5 mm were independently associated with eERP-PDD failure.

Conclusions: EUS-PDD demonstrated higher technical and clinical success than eERP-PDD for PJS, with comparable safety after adjustment. An MPD diameter ≥ 5 mm was associated with eERP-PDD failure. An MPD-based algorithm is proposed: eERP-PDD for MPD < 5 mm with EUS-PDD as salvage, and EUS-PDD for MPD ≥ 5 mm. This algorithm is hypothesis-generating and requires prospective validation.

Objectives: Sedation during colonoscopy is becoming increasingly important. Remimazolam, an ultra-short-acting benzodiazepine, has a shorter pharmacokinetic half-life than that of midazolam. This study examined whether remimazolam provides superior sedation during colonoscopy in Japanese patients.

Methods: This prospective, multicenter, randomized, single-blind, controlled trial included adults (18-80 years) scheduled for sedated colonoscopy. Participants were randomized to the remimazolam and midazolam groups. The primary outcome was the proportion of ambulatory patients 5 min after colonoscopy. Secondary outcomes were successful pre-procedure sedation (Modified Observer's Assessment of Alertness/Sedation [MOAA/S] ≤ 4), recovery time, total sedative dose, and adverse events.

Results: Forty patients were enrolled and analyzed (remimazolam, n = 19; midazolam, n = 21). At 5 min post-colonoscopy, ambulation was achieved in 100% (19/19) of remimazolam patients and 19.1% (4/21) of midazolam patients (p < 0.0001). The median time [interquartile range (IQR)] from procedure end to full alertness (MOAA/S = 5) was 0 [0-0] min for remimazolam and 10 [5-20] min for midazolam (p < 0.0001). The median time [IQR] from procedure end to independent ambulation was 0 [0-5] min for remimazolam and 20 [10-30] min for midazolam (p < 0.001). Pre-procedure sedation was successful (MOAA/S ≤ 4) in 100% of both groups. The median amount [IQR] of total sedative dose was 5 [4-6] mg for remimazolam and 3 [3] mg for midazolam. Hypoxemia occurred in 5.3% and 9.5% of patients in the remimazolam and midazolam groups, respectively.

Conclusions: Compared with midazolam, remimazolam resulted in significantly faster recovery after colonoscopy in Japanese patients, with comparable achievement of target sedation and a low incidence of hypoxemia.

Clinical registration: Trial number: jRCTs071240062.