Search in the Abstract Database

Abstracts Search 2019

P194 Automated real-time endoscopic scoring based on machine learning in ulcerative colitis: Red Density reliability and responsiveness study

P. Bossuyt*1,2, S. Vermeire1, M. Ferrante1, T. Makino3, G. De Hertogh4, R. Bisschops1

1Department of Gastroenterology and Hepatology, University Hospitals Leuven, Catholic University of Leuven, Leuven, Belgium, 2Imelda General Hospital, Department of Gastroenterology, Bonheiden, Belgium, 3Pentax Medical, Product Development Department, Tokyo, Japan, 4 Department of Pathology, University Hospitals Leuven, Catholic University of Leuven, Leuven, Belgium

Background

Endoscopic scoring in ulcerative colitis (UC) is subjective and has poor correlation with histological scoring. Histological remission predicts favourable long-term outcome in UC. Operator-independent automated digital scoring of endoscopic and histological inflammation in UC could provide an objective and predictive evaluation of remission. The aim of this study was to test the operating properties of the Red Density (RD) score (responsiveness and reliability).

Methods

The RD system uses machine learning (ML) to calculate a score based on real-time automatic extraction of pixel data from endoscopic images. This ML algorithm incorporates colour data and vascular pattern recognition. In this prospective study, consecutive patients with UC presenting at the IBD outpatient clinic with symptoms suggestive of a flare were included. At baseline and 8–14 weeks after treatment escalation we recorded endoscopic (Red Density score, Ulcerative colitis endoscopic index of severity [UCEIS], Mayo endoscopic subscore [MES]), clinical (total Mayo, PRO-2), histological data (Robarts histological index [RHI], Geboes score) and C-reactive protein. Investigators were blinded for the RD score. Correlation was tested between RD and clinical, biochemical, endoscopic, and histological scores (Spearman’s rank correlation). Responsiveness was significant if standard effect size >0.8.

Results

Ten patients had two consecutive visits (M/F 4/6, median age 39 years IQR 36–48). At baseline all patients had active endoscopic disease (median (IQR) UCEIS 4.5 (2.5–5); MES 2 (1.3–2)). Nine patients had a change in their endoscopic score after treatment compared with baseline. The median delta in UCEIS and MES was 3 (IQR 2–4) (p = 0.009) and 1 (IQR 1–2) (p = 0.008), respectively. A significant number of patients achieved clinical, endoscopic and histological remission after treatment (all p < 0.03). Median RD score decreased significantly from baseline (166 to 58; p = 0.01) (Figure 1). RD correlated moderate with clinical outcomes (r >0.65, p = 0.001), and strong with both endoscopic (r > 0.75, p < 0.0001), and histological scores (r > 0.75, p < 0.0001). The standardised effect size for RD was 1.22.

Figure 1. Evolution histological and Red Density score. RHI, Robarts histological index; RD, Red Density; w, week.

Conclusion

The automated digital endoscopic Red Density score correlates strongly with endoscopic, histological scores in UC. Red Density demonstrates an excellent sensitivity to change after treatment escalation. Red Density is an ideal operator-independent digital tool for the evaluation of endoscopic and histological disease activity in UC.