Real-time use of artificial intelligence (CAD EYE) in the colorectal cancer surveillance of Lynch syndrome patients - an international multicenter trial

Colorectal cancer (CRC) is one of the most common cancers, with ~61.000 new
diagnoses and over 25.000 deaths per year in Germany. Lynch syndrome (LS) is
the most common hereditary colorectal cancer syndrome and accounts for ~3% of
all CRCs. This autosomal dominant disorder is caused by germline mutations in
DNA mismatch repair genes (MLH1, MSH2, MSH6, PMS2 and EPCAM). One in 279
individuals of the general population is a mutation carrier. Thus, almost
300.000 individuals are expected to have LS in Germany. Carriers of pathogenic
variants are at high risk of CRC with a cumulative incidence of up to 70% by
the age of 70 despite regular endoscopic surveillance, and also at elevated
risk to develop metachronous CRC. LS also includes a variety of extracolonic
malignancies. The number of patients identified with LS is expected to rise in
the next years due to the increasing use of universal screening for mismatch
repair deficiency in solid tumors.

For patients with LS it has been suggested to undergo surveillance colonoscopy
every 1 to 2 years. The ade-noma-carcinoma sequence is considered a major
pathway in the carcinogenesis of CRC also in Lynch syndrome. This opens a
window of opportunity for endoscopic surveillance with removal of adenomatous
polyps. It has been shown that advanced histology (high-grade dysplasia;
villous component) was already apparent in adenomas with a size of 2 - 5 mm.
Thus, it is of paramount importance to endoscopically detect and remove
adenomas as early and sensitive as possible before they can develop into
invasive CRC.

Adenomas are considered to represent the main precursors of colorectal cancer
in Lynch syndrome. Due to the accelerated progression from adenoma to CRC in
LS, high ADR is particularly important in these patients to minimize the risk
of carcinoma development. Accordingly, intensified surveillance strategies with
colonos-copies every one to two years have been shown to reduce both CRC
incidence and CRC-associated mortality. However, several studies have shown
that a relevant proportion of adenomas are missed even when examinations are
performed by experienced endoscopists. This is especially true for flat
lesions, which are typical for LS. These missed adenomas are considered a
possible reason why patients with Lynch syn-drome are at very high risk for
post-colonoscopy CRC with a cumulative 10-year incidence of up to 8% despite
endoscopic surveillance. Furthermore, small colorectal polyps in Lynch syndrome
patients have a higher risk of harboring cancer or high-grade dysplasia
compared to an average-risk population.

Epidemiological studies have reported that the cumulative CRC rate at 70 years
among individuals with LS undergoing colonoscopy surveillance can be as high as
46% among MLH1, 35% among MSH2, 20% among MSH6, and 10% among PMS2 pathogenic
mutation carriers. Some authors postulate that some post-colonoscopy CRCs in LS
may emerge from MMR-deficient crypt foci without a polypoid growth pattern and
therefore can be difficult to detect by colonoscopy. However, retrospective
descriptive studies evaluating post-colonoscopy CRC showed associations with
incomplete examination, inadequate bowel preparation, and possible incomplete
resection of lesions.Furthermore, high miss rates for colorectal neoplasia (12%-
74 %) have been reported in several back-to-back colonoscopy studies.
Therefore, it can be concluded that high quality standards for colonoscopy are
not always met in individuals with LS. However, the evidence regarding key
performance indicators for colonos-copy in individuals with LS is limited.
Almost one third (28%) of these small adenomas are missed according to a
recently published meta-analysis. Often the non-polypoid shape hampers
endoscopic detection.

Several approaches have been made to enhance endoscopic detection of adenomas
by using new endoscopic techniques. Conventional chromoendoscopy is conducted
by the use of indigo carmine as pan-colonic dye applied via spray catheter.
This was studied in several back-to-back endoscopic studies as well as two
ran-domized controlled trial comparing chromoendoscopy to standard white-light
endoscopy. The by far largest and well-designed trial by Rivero-Sánchez et al
showed that high-definition white-light endoscopy is not inferior to
chromoendoscopy. The European Society for Gastrointestinal Endoscopy (ESGE)
states that *the use of chromoendoscopy may be of benefit in individuals with
Lynch syndrome undergoing colonoscopy; routine use must be balanced against
costs, training, and practical considerations*. In individual patient data
meta-analysis of studies comparing chromoendoscopy with WLE endoscopy in Lynch
syndrome for ade-noma detection, chromoendoscopy showed no apparent increase in
adenoma detection.

Several trials have studied the effect of virtual chromoendoscopy
(Narrow-Band-Imaging, Fuji Intelligent Color Enhancement,
Linked-Colour-Imaging) on adenoma or polyp detection. The most recent trial by
Houwen et al. showed no effect on polyp detection as the primary endpoint but
on adenoma detection rate as a secondary endpoint.

Limitations in human visual perception and other human biases such as fatigue,
distraction, level of attention during examination increase such detection
errors. A possible way to mitigate these could be another key to improve polyp
detection and further reduce CRC incidence. The use of artificial intelligence
might be a prom-ising approach in this high-risk patient cohort. In recent
years, several clinical trials in the general population have demonstrated that
AI-assisted colonoscopy is a promising approach, showing significant
improvement in the detection of polyps and adenomas compared with standard
white light endoscopy. These auto-mated intelligence systems can be easily
activated via button on the endoscope. It analyzes the colonic mu-cosa
real-time during the examination and signals the detection of a polyp by an
optical and also acoustic signal.

Fujifilm has developed a technology known as *CAD EYE* to support colonic polyp
detection and characteri-zation during colonoscopy, utilizing Fujifilm's
medical AI technology named REiLI (CADE-EYE, Fujifilm, Japan). CAD EYE is a
customized detection and characterization support compatible with the ELUXEO
and ELUXEO Lite system. CAD EYE Detection is activated when the clinicians are
observing in White Light Mode or LCI Mode. When a suspicious polyp is detected
within the endoscopic image, a Detection Box indicates the area where the
suspicious polyp has been detected accompanied by a sound signal. Furthermore,
it is also able to distinguish polyps between *neoplastic* or *hyperplastic* by
their macroscopic appearance.

We have recently conducted a randomized controlled pilot trial, where Lynch
syndrome patients were random-ized to their regular surveillance colonoscopy
either with or without the use of CADe. In the HD-WLE arm, adenomas were
detected in 12/46 patients compared to 18/50 in the AI arm (26.1% [95% CI
14.3-41.1] vs. 36.0% [22.9-50.8]; p=0.379). In this study, we now want to
confirm these results that the adenoma detection rate (ADR) using CADe is
higher than using HD-WLE.

Primary Objective:
- To compare adenoma detection rate (ADR) with versus without real-time
automated detection (CAD EYE, Fujifilm)

Secondary Objectives:
- To compare the size, morphology and histology of adenomas detected and
resected in both arms of the study in terms of mean/absolute num-ber and
anatomical localisation (proximal/distal)
- To compare the size, morphology and histology of advanced adeno-mas detected
and resected in both arms of the study in terms of mean/absolute number,
detection rate and anatomical localisation (proximal/distal)
- To compare the size, morphology and histology of serrated polyps (sessile
serrated lesions, hyperplastic polyps, traditional serrated ade-nomas) detected
and resected in both arms of the study in terms of mean/absolute number,
detection rate and anatomical localisation (proximal/distal)
- To compare the size, morphology and histology of polyps detected and resected
in both arms of the study in terms of mean/absolute number, detection rate and
anatomical localisation (proximal/distal)
- To compare detection rate and mean/absolute number of adenomas and serrated
polyps (sessile serrated lesions, hyperplastic polyps, tra-ditional serrated
adenomas) with and without CAD EYE depending on center/previous experience
using artificial intelligence systems
- To compare detection of colorectal cancer with and without CAD EYE
- To compare examination/procedure times with and without CAD EYE
- To assess risks of formation of sessile serrated lesions/adenoma/carci-noma
formation depending on the interval to the last colonoscopy as well as on the
findings of the previous examination and underlying pathogenic variant
- To assess specificity/sensitivity of the polyp differentiation mode of the
CAD EYE system
- To explore false positives lesions by the CAD EYE system

The study is a prospective, international, multicenter, randomized, controlled,
open-label, two-arm study carried out by experienced endoscopists in expert
centers. Once a subject is determined to be eligible for the study, the subject
will be randomized to one of the two trial arms. Asymptomatic patients with a
diagnosis of Lynch syndrome will be enrolled in this trial. All subjects will
be randomized to 1 of the 2 treatment groups in a 1:1 ratio as described below.
Randomization will be stratified by previous colorectal cancer (yes or no),
underlying (likely-) pathogenic variant (MLH1, MSH2 (EPCAM), MSH6, PMS2),
center, gender (male/female) and interval to last colonoscopy (12-23 months,
24-36 months, index colonoscopies or >36 months). Randomization arms: - Arm 1:
standard high-definition white light endoscopy (HD-WLE) without real-time use
of artificial intelligence software (CAD EYE) for polyp detection, but with use
of artificial intelligence software for polyp characterization - Arm 2:
standard high-definition white light endoscopy (HD-WLE) with real-time use of
artificial intelligence software (CAD EYE) for both detection and
characterization of colorectal polyps

No additional risk or burden associated with study participation.