P0279Automated bowel wall layer assessment in post-operative Crohn’s Disease: the Prosper study

Abstract Background Up to 75% of CD patients require surgery during disease course, and post-operative recurrence (POR) occurs in up to 70% patients within 1 year. Early, accurate and frequent monitoring is therefore critical for optimal management. Intestinal ultrasound (IUS) has emerged as a safe and widely available tool to assess CD activity. Bowel wall thickness (BWT) is a key feature that correlates with the presence of endoscopic recurrence. However, IUS interpretation is often burdened by subjectivity and high interobserver variability, underscoring the need for standardised, objective assessment tools. Artificial intelligence (AI) offers an opportunity to enhance precise and reproducible IUS-based disease evaluation. We aimed to develop a novel computer-aided model for automated quantitative measurement of BWT across different intestinal segments. Methods IUS videos from CD patients enrolled in the international, multicentre PROSPER study (NCT06505304) were included. 70 high-quality frames from 22 videos were manually and randomly extracted. Additional 12 frames were extracted to develop the computer-aided system to detect bowel wall layers from the cross section of bowel wall. Regions of interest (ileum, anastomosis, and colon) were identified and bowel wall layers were manually segmented as ground truth. Drawing a line across the bowel wall (from lumen to serosa) was required as input. After extracting the grey-level intensity from the frames along this line, the intensity profile was smoothed to reduce noise. Layers’ boundaries were automatically identified by detecting most relevant peaks and valleys in the grey-level intensity profile. Considering the sequence and echogenicity of layers, the thickness of each layer was computed, enabling accurate estimation of overall BWT (Figure 1). The diagnostic performance was evaluated by correlating computer-derived measurements with experts manual assessment. Results Table 1 details model performance. A very high correlation (r 0.98; mean absolute error 3.8 pixels, mean percentual error 8.63%) was identified between the manual and the automated estimation of BWT. Excellent performance was confirmed in assessed intestinal tracts: r 0.97 in the neo-terminal ileum, r 0.98 in the anastomosis and r 0.97 in the colon proximal to the anastomosis. Conclusion Our semi-automated method accurately measure BWT through reliable layer stratification and demonstrates excellent concordance with manual assessment. This approach offers a promising avenue for fully automated identification and quantification of bowel wall layers on IUS. Further refinement and integration of additional sonographic features will enhance standardization and clinical utility in POR monitoring. Conflict of interest: Iacucci, Marietta: No conflict of interest Dr. Zammarchi, Irene: No conflict of interest Verstockt, Bram: No conflict of interest Allocca, Mariangela: Personal Fees: consulting fees from Nikkiso Europe, Mundipharma, Janssen, Abbvie, Pfizer, Ferring, Galapagos, Sandoz, Lilly and Alfasigma Bezzio, Cristina: Personal Fees: I received consulting/advisory board/lecture fees from Alfa Sigma, AbbVie, Celltrion, Eli Lilly, Ferring, Gilead, Johnson & Johnson MSD, Pfizer and Takeda Banai Eran, Hagar: No conflict of interest Castiglione, Fabiana: Honoraria from: Takeda, AbbVie, Celltrion, Johnsson Johnsson, Cadigroup, Sandoz, Pfizer, Lilly, Lionhealth, Nestlè Cannatelli, Rosanna: No conflict Dal Buono, Arianna: speaker’s fees from AbbVie, Alphasigma, Ferring, Lilly, Janssen, and Celltrion Doherty, Glen: No conflict of interest Furfaro, Federica: Grant: IG-IBD Personal Fees: Pfizer, Biogen, J&J, Abbvie, Amgen, Janssen Lepore, Federica: No conflict of interest Lo Bello, Antonio: No conflict of interest Lu, Cathy: No conflict of interest Maconi, Giovanni: No conflict of interest Nardone, Olga Maria: No conflict of interest O’grady, John: No conflict of interest Piazza O Sed, Nicole: No conflict of interest Ricci, Chiara: No conflict of interest Ghosh, Subrata: No conflict of interest Naranjo, Valery: No conflict of interest Grisan, Enrico: No conflict of interest