Artificial intelligence in capsule endoscopy (ArtIC)

  • Research type

    Research Study

  • Full title

    Role of the artificial intelligence in capsule endoscopy (ArtIC) for the identification of small bowel lesions in patients with small intestinal bleeding

  • IRAS ID

    297101

  • Contact name

    Mark McAlindon

  • Contact email

    mark.mcalindon@nhs.net

  • Sponsor organisation

    Sheffield Teaching Hospitals NHS FT

  • Clinicaltrials.gov Identifier

    NCT04821349

  • Duration of Study in the UK

    0 years, 11 months, 30 days

  • Research summary

    Research Summary

    Capsule endoscopy (CE) is a swallowable pill camera, safe, patient friendly and performed for the evaluation of the small bowel which is difficult to reach using conventional endoscopy. The most common indication for CE is suspected small bowel bleeding. The capsule acquires images of the gastrointestinal tract during transit and transmits them to a data recorder which the patient wears on a belt.

    This raw data is converted to a video format on a computer which the clinician can view. Meta-analysis shows that CE has a diagnostic yield of about 60% for this indication, and angiodysplasias are the most relevant findings, accounting for 50%. It is now an accepted first line small bowel examination in suspected bleeding if conventional upper and lower endoscopy do not reveal the cause. The main drawback of CE is the time and concentration needed for the clinician to view the video and produce the report, a process which may take 30-60 minutes, per case. In order to reduce reading time, several softwares have been developed which recognise and remove sequentially identical (or near identical) images. These have been shown to significantly reduce reading time without impacting the miss rate in diffuse mucosal disease (such as Crohn’s disease). This may not be the same with isolated lesions since several studies report an unacceptable miss rate.

    New advances such as artificial intelligence are being investigated. Deep convolutional neural networks (CNNs) are algorithms developed by the computer analysis of thousands of pathological images to establish an extensive and complex range of recognition features used to identify images containing potential abnormalities in clinical practice.

    The aim of the study is to determine the role of AI-assisted CE reading in clinical practice.

    Summary of Results

    Artificial intelligence (AI) software was compared to expert human readers of small bowel capsule endoscopy videos. AI-assisted reading proved superior to human reading, identifying more lesions causing anaemia, and with a 9-fold quicker reading time.

  • REC name

    London - Chelsea Research Ethics Committee

  • REC reference

    21/PR/0711

  • Date of REC Opinion

    8 Jun 2021

  • REC opinion

    Further Information Favourable Opinion