Introduction

In recent years, there has been a surge in the development of artificial intelligence (AI) systems for endoscopic applications, such as Barrett’s neoplasia detection and polyp characterization. While these systems have shown promising results, the vast majority is trained and validated on small, retrospective, single-center or expert-acquired data sets. Current AI systems are particularly prone to a so-called domain-shift: a system only performs well on data it has been trained on. As real-world scenarios in the endoscopy suite can be complex and diverse, the risk of poor generalization of AI should not be underestimated.

Aims & Methods

In this study we aim to provide a quantitative analysis of this domain shift and propose several strategies to improve generalization to clinical practice. We trained three AI systems for commonly used applications in gastrointestinal endoscopy, using publicly available data sets with high quality standards. In order to identify the limitations of these systems, all test sets were artificially degraded in a stepwise manner with clinically prevalent artifacts: blur, inadequate exposure and color variability. These degradations represent the heterogeneity of image quality in daily clinical practice. Severity of degradation was clinically calibrated by a panel of expert endoscopists. Two levels of degradation were identified: 1) mild degradation, still considered as high quality data; 2) moderate degradation, considered as image quality that could still be encountered in daily clinical practice. The trained models were subsequently evaluated for their appropriate performance metrics on the gradually degraded test sets. After implementation of several training strategies to mitigate potential performance loss, the models were re-evaluated on the same test sets.

Results

All three models display a considerable decrease in performance when exposed to various types of clinically relevant image degradation. Mean loss was 7.4% and 16.6% for respective mild and moderate degradation. Implementation of specific training strategies to improve robustness resulted in substantial reduction of performance loss on all degraded test sets.

Conclusion

This study highlights the potential limitations of current AI systems in endoscopic applications which may lead to poor generalization. Specific training strategies can mitigate this risk and improve robustness of these systems. Development of robust and reliable AI systems that deliver consistent and accurate performance is crucial to ensure safe implementation into daily clinical practice.

Introduction

The LIR!C trial demonstrated that laparoscopic ileocaecal resection (ICR) is a cost-effective alternative to infliximab therapy (IFX) in patients with uncomplicated ileocaecal Crohn’s disease (CD), with comparable quality of life outcomes. The aim of this study was to assess the 10-year follow-up of patients enrolled in the LIR!C trial.

Aims & Methods

This retrospective follow-up study included patients from the original LIR!C trial, a multicentre randomized controlled trial comparing ICR to IFX treatment in patients with ileal CD. Long term follow-up data were collected between February 3 and April 14, 2025. Data extraction was focused on initiation or switch of IBD treatments (corticosteroids, immunomodulators and/or advanced medical therapies) and (re)-resections. The follow-up period ranged from trial enrolment to the last visit with the gastroenterologist or surgeon. Primary outcomes were: (1) proportion of patients in clinical remission without therapy switch (no medical therapy or re-resection in the ICR group, and continuation or the cessation of IFX in case of remission in the IFX group), and (2) therapy-free remission, defined as clinical remission without IBD medication and (re)resections.

Results

Follow-up data were available from 60 ICR patients and 57 IFX patients (87% of 134 patients originally included). The median follow-up time was 12 years (IQR 9-14). The clinical remission rate without therapy switch was 40% in the ICR group compared to 26.3% in the IFX group (p=0.117). Of the 60 ICR patients, 24 (40%) had no additional therapy, 13 (21.7%) received an immunomodulator, 18 (30%) biological therapy and 5 (8.3%) underwent a second resection after biological treatment. In the IFX group, 33 (57.9%) required an ICR of whom 25 (75.8%) had previously received second-line biological treatment. Additionally, 7/57 patients (12.3%) were therapy free, 8 (14%) continued IFX treatment, 6 (10.5%) received other biological treatment, and 3 (5.3%) switched to an immunomodulator. Therapy-free remission was significantly higher in the ICR group when compared to the IFX group (40% vs 12%, p < 0.001).

Conclusion

After 10 years of follow-up, the clinical remission rate without therapy switch was comparable between both groups, highlighting the long-term durability of both treatment approaches. The majority of IFX patients needed ICR within 10 years of follow-up. 40% of the ICR group did not require any additional therapy further supporting ICR as first-line treatment option in ileal CD.

References

Ponsioen, C. Y., de Groof, E. J., Eshuis, E. J., Gardenbroek, T. J., Bossuyt, P. M. M., Hart, A., Warusavitarne, J., Buskens, C. J., van Bodegraven, A. A., Brink, M. A., Consten, E. C. J., van Wagensveld, B. A., Rijk, M. C. M., Crolla, R. M. P. H., Noomen, C. G., Houdijk, A. P. J., Mallant, R. C., Boom, M., Marsman, W. A., Stockmann, H. B., … LIR!C study group (2017). Laparoscopic ileocaecal resection versus infliximab for terminal ileitis in Crohn's disease: a randomised controlled, open-label, multicentre trial. The lancet. Gastroenterology & hepatology, 2(11), 785–792. https://doi.org/10.1016/S2468-1253(17)30248-0

Disclosure

GD'H served as a consultant for AbbVie, Alimentiv, AstraZeneca, Bristol Myers Squibb, Celltrion, Eli Lilly, Exeliom Biosciences, Johnson & Johnson, Pfizer, and Takeda; and has received speakers bureau fees from AbbVie, Eli Lilly, Pfizer, Bristol Myers Squibb, and Takeda.

Introduction

To evaluate the utility of the unroofing technique in the diagnosis of gastrointestinal subepithelial tumors (SETs), we performed a multicenter retrospective study.

Aims & Methods

This study was conducted in 10 hospitals and involved all eligible patients who underwent unroofing techniques to gain biopsy for gastrointestinal SETs between April 2015 and March 2021. The primary endpoint was the rate of correct diagnosis with the unroofing technique, and the secondary endpoints were the incidence of adverse events and the factors contributing to the correct diagnosis.

Results

The study included 61 patients with 61 gastrointestinal SETs, a median tumor size of 20 mm, and a median procedure time of 38 minutes with 82% successful tumor exposure. The rate of tissue acquisition was 98.4%, and the rate of pathological diagnosis was 72.1%. In 44 cases with a pathological diagnosis, two showed discrepancies with the postresection pathological diagnosis. No factors, including facility experience, organ, tumor size, or tumor exposure, significantly affected the correct diagnosis. No significant differences in the correct diagnosis rate were observed between less than ten experiences and more than ten experiences (74.1 vs. 70.6%). The correct diagnosis rates for tumors <20 mm and ≥20 mm were 76.9% and 65.7%, respectively. There was one case of delayed bleeding and two cases of perforations.

Conclusion

The diagnostic yield of the unroofing technique was acceptable. The unroofing technique was beneficial regardless of institutional experience, organ, tumor size, or actual tumor exposure.

Introduction

Long-term outcomes of gastric subepithelial lesions have not been elucidated. To reveal the natural history, we initiated a prospective, 10-year follow-up of patients with small (≤20 mm) gastric subepithelial lesions in September 2014. Here, we report the results of an interim analysis of a prospective observational study.

Aims & Methods

In total, 567 patients with 609 lesions were prospectively registered between September 2014 and August 2016. The location, size, morphology, and number of subepithelial lesions were recorded on a web-based case report form.

Results

The endoscopic follow-up period was 4.60 ± 1.73 years (mean ± SD), and survival data were investigated for 5.28 ± 1.68 years. This interim analysis revealed that the estimated cumulative incidence of a size increase ≥5 mm, after accounting for patients’ death and resection of the tumor as competing risk events, was 4.5%. In addition, the estimated cumulative incidence of lesion size increase ≥5 mm or resection of lesions was 7.7% at 5 years, and that of size increase ≥10 mm or resection of lesions was 4.3% at 5 years.

Conclusion

These results indicate that approximately one in 13 patients with small (≤20 mm) gastric subepithelial lesions may require resection or further investigation for increased tumor size (≥5 mm) within five years under active surveillance.

References

Nishida T, Hirota S, Yanagisawa A et al. Clinical practice guidelines for gastrointestinal stromal tumor (GIST) in Japan: English version. Int J Clin Oncol 2008; 13: 416-30.

Disclosure

This study has been conducted as an Academic Committee Working Group of the Japan Gastroenterological Endoscopy Society.