Introduction

Bowel preparation quality is a key determinant of colonoscopy effectiveness and adenoma detection^1,2. Manual evaluation of free-text colonoscopy reports is labor-intensive and limits quality monitoring³. Developing a natural language processing (NLP) application for measuring colonoscopy quality demonstrates the potential to automate this process by extracting key quality indicators⁴.

Aims & Methods

We aimed to develop and validate a NLP approach to automatically classify bowel cleanliness and to comprehensively structure colonoscopy reports by extracting procedural and pathological variables. A total of 11,374 reports from our institutional endoscopy unit were analyzed. An annotated dataset of 1,018 reports labeled as “good,” “intermediate,” or “poor” preparation was used to train a TF-IDF–based logistic regression classifier. Cross-validation and test performance were evaluated using accuracy, precision, recall, and F1-scores, and the model was applied to the full dataset. Additional variables (cecal intubation, polyp presence, number, size, localization, polypectomy method, and hemostasis application) were extracted using rule-based NLP heuristics. Adenoma detection rate (ADR) was calculated in a predefined subset (January 2023–July 2025) with complete pathology data, and multivariable logistic regression identified independent predictors.

Results

The NLP classifier achieved a cross-validated accuracy of 92%. Precision and recall were highest for “good” preparation (0.98 and 0.99) and acceptable for “intermediate” (0.85 and 0.82) and “poor” (0.85 and 0.87). Application to the full cohort yielded 5,787 (50.9%) good, 2,176 (19.1%) intermediate, and 3,411 (30.0%) poor preparations. Leveraging these structured data, ADR analysis in the subset of 5,420 patients showed an overall rate of 20.4%. Multivariable analysis demonstrated that ADR was independently associated with older age (OR 1.05 per year, 95% CI 1.05–1.06, p<0.001), male sex (OR 1.74, 95% CI 1.51–2.01, p<0.001), adequate bowel preparation (OR 1.46, 95% CI 1.21–1.75, p<0.001), cecal intubation (OR 1.76, 95% CI 1.43–2.16, p<0.001), and procedures performed by specialists compared with trainees (24.0% vs 20.3%; OR 1.25, 95% CI 1.05–1.47, p=0.010).

Conclusion

This large-scale real-world analysis demonstrates the feasibility of NLP-based automated classification of bowel preparation quality and its direct utility in generating structured colonoscopy datasets for quality monitoring. By integrating bowel cleanliness with other performance indicators (age, sex, cecal intubation, operator expertise), our framework not only achieved high classification accuracy but also provided clinically relevant benchmarks for ADR, supporting targeted quality improvement in endoscopy units.

References

1. Clark BT, Protiva P, Nagar A, et al. Quantification of Adequate Bowel Preparation for Screening or Surveillance Colonoscopy in Men. Gastroenterology. 2016;150(2):396-e15. doi:10.1053/j.gastro.2015.09.041
2. Clark BT, Rustagi T, Laine L. What level of bowel prep quality requires early repeat colonoscopy: systematic review and meta-analysis of the impact of preparation quality on adenoma detection rate. Am J Gastroenterol. 2014;109(11):1714-1724. doi:10.1038/ajg.2014.232
3. Raju GS, Lum PJ, Slack RS, et al. Natural language processing as an alternative to manual reporting of colonoscopy quality metrics. Gastrointest Endosc. 2015;82(3):512-519. doi:10.1016/j.gie.2015.01.049
4. Harkema H, Chapman WW, Saul M, Dellon ES, Schoen RE, Mehrotra A. Developing a natural language processing application for measuring the quality of colonoscopy procedures. J Am Med Inform Assoc. 2011;18 Suppl 1(Suppl 1):i150-i156. doi:10.1136/amiajnl-2011-000431

Introduction

Capsule endoscopy (CE) is a minimally invasive and essential tool for assessing small bowel disorders, especially Crohn’s disease. Accurate detection of ulcers and erosions is vital for evaluating disease activity, guiding treatment, and monitoring therapeutic response. Although CE has revolutionized lesion detection, its manual interpretation remains time-consuming and subject to interobserver variability. Recent advances in artificial intelligence (AI), particularly convolutional neural networks (CNNs), offer promising solutions by enhancing diagnostic accuracy, reducing variability, and improving workflow efficiency. This study aimed to develop and validate a robust AI model capable of detecting and distinguishing small bowel ulcers and erosions across multiple CE platforms and clinical centers.

Aims & Methods

In a prospective, multicenter study conducted between January 2021 and April 2024, data were collected from centers in Portugal, Spain, and the United States. Two CE devices—PillCamSB3 and Olympus EC-10—were used to analyze 137 anonymized exams. The AI-assisted readings generated by a deep learning model were compared to standard-of-care (SoC) interpretations, with an expert consensus panel serving as the reference standard. Performance metrics included sensitivity, specificity, positive and negative predictive values (PPV, NPV), and area under the receiver operating characteristic curve (AUROC).

Results

Expert consensus identified ulcers and erosions in 56 patients (40.9%). Compared to the reference standard, SoC readings achieved 60.7% sensitivity, 98.8% specificity, 97.1% PPV, 78.4% NPV, and 83.2% accuracy. In contrast, the AI-assisted model achieved 94.6% sensitivity, 80.2% specificity, 76.8% PPV, 95.6% NPV, and 86.1% accuracy. The AI model was found to be both non-inferior (p < 0.001) and statistically superior (p < 0.001) to SoC. Notably, it identified 68 lesions, nearly double those detected by SoC (35), and demonstrated consistent performance across devices and centers. The mean AI-assisted reading time was 239 seconds (SD 138 seconds) per exam.

Conclusion

The AI-assisted model significantly outperformed conventional SoC in the detection of small bowel ulcers and erosions, achieving higher sensitivity and overall accuracy while substantially reducing reading time. This is the first multicenter validation of an interoperable AI solution for CE, capable of high diagnostic performance across different devices and clinical environments. These findings highlight the potential of AI to transform endoscopic evaluation and management in inflammatory bowel disease.

Introduction

Endoscopic Submucosal Dissection (ESD) is the best endoscopic technique that enables en bloc resection of superficial colorectal neoplasms.Despite its effectiveness, ESD is not very widespread in Western countries due to technical complexity, long procedure times, and a steep learning curve. These barriers have created a significant gap in adoption and outcomes between Eastern and Western centers. Underwater ESD (U-ESD) has been introduced to simplify the procedure. This study aimed to evaluate the effectiveness of U-ESD versus conventional gas-based ESD (G-ESD) in a Western center, focusing on procedural efficiency and safety.

Aims & Methods

This is a single-center, prospective randomized controlled trial conducted at Fondazione Campus Bio-Medico in Rome between December 2024 and April 2025.Forty patients with non-pedunculated colorectal lesions were randomly assigned in a 1:1 ratio to undergo either Underwater ESD (U-ESD) or conventional ESD with insufflation of Carbon Dioxide (CO2) (G-ESD).The primary outcome was total procedure time. Secondary outcomes included dissection speed, en bloc and R0 resection rates, use of hemostatic forceps, and the incidence of adverse events such as delayed bleeding, perforation, and post-ESD coagulation syndrome (PECS).

Results

A total of 40 patients referred to our endoscopic unit for ESD between December 2024 and April 2025. The mean size of the colorectal lesions was 57.4 mm and 40.7 mm in the U-ESD and G-ESD groups. The location of the lesions was in the rectum in 15% and 55% of the U-ESD and C-ESD groups, respectively; in the right side of the colon in 35% and 20% of the U-ESD and G-ESD groups, respectively. The morphology of LST was classified as LST-nongranular in 20 % and 25% of the U-ESD and G-ESD groups, respectively; as LST granular in 60% and 35% of the U-ESD and G-ESD groups, respectively. Fibrosis (grade 1 and 2) was present in 45% and 30% of the U-ESD and G-ESD groups, respectively. The mean total procedure time was shorter in the U-ESD group compared with the G-ESD group (53.3 ±33 minutes vs 78.7 ± 25.6 minutes; mean difference,–25.4 minutes; P=0.009).Mean dissection speed was faster in the U-ESD group compared with the G-ESD group (42 ± 26.3 mm2/min vs 13.3 ± 8 mm2/min; mean difference, 28.7 mm2/min; P= 0.001). En bloc and R0 resection rates in both groups were 100%. The use of a hemostatic forceps was significantly lower in U-ESD compared G-ESD (40% vs 90%, P= 0.001). Delayed bleeding occurred in only one patient in G-ESD group and was managed successfully by endoscopic clipping. No perforations occurred in either group. PECS was noted in only one patient in the G-ESD group.

Conclusion

UESD has shown a significantly shorter procedure time and faster dissection speed compared with GESD. These both empower the possibility of considering U-ESD as the preferred colorectal ESD method for colorectal lesions.