Comparing the Number of False Activations Between Two Artificial Intelligence CADe Systems: the NOISE Study

NCT ID: NCT04399590

Last Updated: 2021-09-16

Basic Information

• Recruitment Status: COMPLETED
• Total Enrollment: 40 participants
• Study Classification: OBSERVATIONAL
• Study Start Date: 2020-09-01
• Study Completion Date: 2021-03-31

Brief Summary

One fourth of colorectal neoplasias are missed during screening colonoscopies-these can develop into colorectal cancer (CRC). In the last couple of years, Artificial Intelligence Deep learning systems were introduced in the endoscopic setting to allow for real-time computer-aided detection/characterization (CAD) of polyps with high- accuracy. Few CADe (detection) and CADx (diagnosis, characterization) have been therefore proposed with this purpose. Because CAD systems are based on deep learning where the computer directly learns polyp recognition from supervised data without any human-control on the final algorithm, their outcome incorporates some unpredictability in the clinical setting that must be cautiously interpreted after its application. This means that the endoscopist may be presented with FP images that he would have never been selected in the first place as suspicion areas. These FPs may hamper the efficiency of CADe-colonoscopy. Additional time may be required to discriminate between an actual FP and a possible false negative result. An excess of FPs may reduce the motivation of the endoscopist for CADe, leading to its underuse in clinical practice. Although the indications of a CADe must always be interpreted by physician, FP may result in unnecessary polypectomy with related adverse events when used without appropriate training. Yet, there is a lack of information among quantity and quality of False Positive signals provided by the systems. From a post-hoc analysis of a Randomized Clinical Trial, in which we extracted and analysed a video library of CADe-colonoscopy (GI Genius) performed in our institution Humanitas Clinical and Research Hospital IRCCS we aimed that False positives by CADe are primarily due to artefacts from the bowel wall. Despite a high frequency, FPs from this CADe system resulted in a negligible 1% increase of the total withdrawal time as most of them were immediately discarded by the endoscopists.

Conditions

Artificial Intelligence

Study Design

• Observational Model Type: COHORT
• Study Time Perspective: PROSPECTIVE

Interventions

Interficial Intelligence

Interficial Intelligence

Intervention Type: OTHER

Eligibility Criteria

Inclusion Criteria

1. Age over 18 years

2. Ability to provide and to give informed consent

3. Boston Bowel Preparation Score > 6 (>2 each segment)

Exclusion Criteria

1. Boston Bowel Preparation Score < 6 (<2 each segment)

2. Patients who had chronic inflammatory bowel diseases (such as Chron or Ulcerative Colitis)

3. Inability to obtain written informed consent

4. Patient unwilling to participate to the study

• Minimum Eligible Age: 18 Years
• Eligible Sex: ALL
• Accepts Healthy Volunteers: No

Sponsors

Istituto Clinico Humanitas

OTHER

Sponsor Role: lead

Responsible Party

Responsibility Role: SPONSOR

Locations

Endoscopy Unit, Humanitas Research Hospital

Rozzano, Milano, Italy

Countries

Italy

References

Spadaccini M, Hassan C, Alfarone L, Da Rio L, Maselli R, Carrara S, Galtieri PA, Pellegatta G, Fugazza A, Koleth G, Emmanuel J, Anderloni A, Mori Y, Wallace MB, Sharma P, Repici A. Comparing the number and relevance of false activations between 2 artificial intelligence computer-aided detection systems: the NOISE study. Gastrointest Endosc. 2022 May;95(5):975-981.e1. doi: 10.1016/j.gie.2021.12.031. Epub 2022 Jan 4.

Reference Type: DERIVED

PMID: 34995639 (View on PubMed)

Other Identifiers

2598

Identifier Type: -

Identifier Source: org_study_id