TBA (22W127)

Capsule endoscopy with artificial intelligence assisted technology: AI in clinical practice.

Author(s)

F O'Hara, A O'Connor, S.O'Donnell, N.Breslin, B.Ryan, D.McNamara

Department(s)/Institutions

Tallaght University Hospital TAGG, Department of Medicine, Trinity College Dublin

Introduction

Capsule endoscopy (CE) reading is a time-consuming process with reading times ranging between 30–120 minutes. Artificial intelligence (AI) in CE is an attractive solution for reducing reading time by removing redundant images and simplifying the identification of abnormalities. Previous studies have demonstrated impressive sensitivity and specificity in created datasets of capsule images, but real-world data is lacking. OMOM® HD Capsule includes Smartscan technology which includes redundancy deletion, lesion detection, and classification software.

Aims/Background

Our aim was to evaluate the OMOM HD Smartscan software in a real-world setting against experienced capsule readers.

Method

OMOM® HD Capsule was employed prospectively in unselected patients presenting for CE. Recordings were then read by 2 separate modalities; Standard reading mode (SR) and “Smartscan mode” (SS) which was read after image processing by AI-assisted technology.

Results

40 patient procedures were included for analysis, mean age of 50 years (18-74); 65% male. All patients completed the procedure uneventfully. Complete visualization of the SB was achieved in 97.5% (n =39) of patients. No capsule retention was recorded. The indications for procedure were occult gastrointestinal bleeding 47.5% (n=19), suspected Crohn’s disease 37.5% (n=15) and Crohn’s disease assessment in 7.5% (n=3) The average reading time was significantly shorter using SS (2.4min) vs SR (30.3min), p < 0.001). There was excellent agreement between both modalities for lesion detection with 100% correlation for positive and negative studies. While a per lesion analysis (n = 373) also showed excellent correlation (k = 0.996). The only lesions not recorded by SS were 2 small angiodysplasias and a circumferential ulcer in an area of poor prep.

Conclusions

The correlation between SS and SR for the detection of pathology is excellent. SS has the potential to significantly improve reading time in CE without negatively affecting diagnostic yield.