Comparison Between Artificial Intelligence and Standard Reading to Investigate Suspected Crohn Disease: the SCAI STUDY

Overview

The diagnosis of Crohn's Disease (CD) is based on a combination of clinical, biochemical (serological and fecal), endoscopic, radiological, and histological investigations. In the absence of obstructive symptoms or known stenosis, European guidelines recommend to investigate the small intestine using Video Capsule Endoscopy (VCE) if ileocolonoscopy is not decisive. To reduce the reading time of VCE and increase the number of identified lesions during the examination, various artificial intelligence software/tools have been developed in recent decades. This study aims to be the first prospective multicentric real-life trial to evaluate AI-assisted VCE using SmartScan in identifying typical mucosal abnormalities of the small intestine in patients with suspected CD and its ability to reduce reading time while maintaining the same diagnostic yield and diagnostic accuracy of standard reading. The objective of the study is to evaluate the role of AI-assisted VCE using the OMOM SmartScan in detecting typical small bowel inflammatory lesions (i.e. erosions and ulcers) in patients with suspected CD, and comparing AI with standard reading.

SCREENING Patients with signs and symptoms of suspected CD AND negative/inconclusive ileocolonoscopy will be enrolled. Inconclusive ileocolonoscopy will be defined as the presence of aspecific inflammatory mucosal alterations in the ileum or colon, that do not allow a final diagnosis of CD. At the Screening Visit, active or recent gastrointestinal infection will be ruled out. Patients who will present with sub-occlusive symptoms will undergo a patency capsule to exclude the presence of a stenosis. Only patients with a prompt patency transit (expulsion within 30-33 hours of the intact patency capsule or its absence on abdominal X-ray) will be enrolled. Signs and symptoms of suspected CD will be defined according to the ECCO Guidelines and ICCE criteria. An ad-hoc protocol defined by the authors will be applied to define the patient with suspicion of CD, stratifying all factors as major or minor criteria: the presence of at least two major criteria or the combination of one major criteria with three minor criteria will be necessary for patient enrolment. Other clinical, laboratory and radiological data will be collected (i.e. Thrombocyte, leukocytes, ESR, Vitamin D). Before the ingestion of the capsule, patients reported outcomes (PRO) measures will be collected to investigate how patients' symptoms impair their physical and mental health (anxiety, depression), interfere with daily activities (work, personal care) and are responsible for chronic fatigue and tiredness, or loss of energy. The aforementioned questionnaires are FACIT-F (https://www.facit.org/measures/facit-fatigue), SF 36 (22), and EQ-5D-5L (23). VCE PROTOCOL: Small bowel VCE will be performed using the OMOM Capsule System (Chongqing Jinshan Science \& Technology (Group) Co), equipped with a DNN based system called SmartScan (SC), which is able to automatically select suspected lesions thus creating a very short video constituted only by selected images. The SC100 system tested in this study consists of: 1. Ingestible microcamera SC1 OMOM HD: the capsule is 25.4 x 11.4 mm in size, with a weight of 3.0 grams. It is embedded with a single camera with a 172° field of view; sampling rate varies between 4 and 35 frames per second, and the image resolution is 512 x 512 pixels. The depth of field is optimized between 0 and 50 mm. The Truecolor 24-bit model allows for high colour depth, displaying up to 16 million colours per pixel. 2. Recorder and SC-RD1 belt: antennas receive data from the OMOM video capsule 3. Vue Smart Work Station: SmartView reading software equipped with the CNN model. The CNN model analyses raw data during the download phase from the recorder and during video review. It supports the operator in reviewing and reporting through three functionalities: * SmartScan: Based on an AI algorithm developed and validated for reviewing capsule videos of the stomach and small intestine, it eliminates redundant images and identifies 16 different types of abnormalities found in the images. * SmartView: Allows the operator to quickly "navigate" through the entire video, showing only the images selected by SmartScan and highlighting the selected video portions on the timeline that can be viewed separately from the rest of the footage. * SmartFinding: Displays only the images selected by the operator, providing a description and characterization of the detected abnormalities. VCE will be performed at each centre according to local regulations and requirements, and the study protocol will address the bowel preparation regimen as well as the reading methods and post-procedural analysis for each patient. PRE-PROCEDURAL PATIENT MANAGEMENT: To achieve homogeneous results, a standard bowel preparation regimen will be adopted, involving the intake of 1 L of Polyethylene glycol + ascorbic acid (Moviprep) in a single-dose manner after the VCE ingestion, as recommended by Xavier et Al. and Estevinho et al. POST-PROCEDURAL MANAGEMENT: Patient follow-up will occur according to routine clinical practice, collecting any cases of retention or adverse events. Patients undergoing VCE will be instructed to check for the expulsion of the capsule and to report any relevant symptom that may suggest potential capsule retention. Patients who do not notice the expulsion of the capsule within 2 weeks of ingestion will undergo an abdominal X-ray to exclude capsule retention. Clinical and laboratory data of all patients enrolled will be collected at follow-up visits (at 6 and 12 months). VIDEO CAPSULE READING: At the site of patient's enrolment, investigators (with experience of \> 100 capsules read) will perform VCE by evaluating the video in standard mode according to the recommendations of the European Society of Gastrointestinal Endoscopy, at 10 frames per second in the small intestine, and 20 frames per second in the esophagus, stomach, and colon. The cleanliness level will be assessed as adequate (excellent/good) or inadequate (fair/poor) according to the qualitative scale of Brotz. Inflammatory lesions will be described using the Nomenclature and semantic descriptions defined by Leenhardt et al. in the International Delphi consensus, specifically modified for this study to ensure inter-observer agreement in the description of lesions. Lesions will be reported specifying their time of appearance (h: min: s, timing of single frames. At the end of the reading, readers will also calculate the Lewis score for each video. A score between ≥ 135 and \< 790 indicates mild inflammation; and a score of ≥ 790 signifies moderate-to-severe inflammation. Investigators at the enrolment centre will anonymize the video which will be randomly reallocated to another centre for the second blind reading assisted by AI (i.e., the readers performing the capsule reading in AI mode will not be aware of the results of the first reading in standard mode). An external expert reader (read more than 500 capsules) will review all the VCE, blind to the source of the videos.