Colonic Polyp
Conditions
Keywords
Deep Learning, Polyp detection rate, CNN
Brief summary
The objective of this study is to compare the polyp detection rate (PDR) of endoscopists unaware of a commercially available artificial intelligence (AI) device for polyp detection during colonoscopy and the PDR of endoscopists with the aid of such a device. Moreover, an extensive characterization of the performance of this device will be done.
Detailed description
Recently, there have been remarkable breakthroughs in the introduction of deep learning techniques, especially convolutional neural networks (CNNs), in assisting clinical diagnosis in different medical fields. One of these artificial intelligence (AI) devices to diagnose colon polyps during colonoscopy was launched in October 2019. Its intended use is to work as an adjunct to the endoscopist during a colonoscopy with the purpose of highlighting regions with visual characteristics consistent with different types of mucosal abnormalities. It is essential to know whether deep learning algorithms can really help endoscopists during colonoscopies. Several studies have already addressed this issue with different approaches and results. However, one common drawback of these type of Machine vs Human retrospective studies is endoscopist bias. It is usually generated because of human natural competitive spirit against machine or human relaxation because of AI-reliance. This can have an effect in the overall results. The investigators perfomed colonoscopies with the use of a commercially available AI system to detect colonic polyps and recorded them during clinical routine. Additionally from March 2019 - May 2019, 120 colonoscopy videos were performed and captured prospectively without the use of AI. In this study, the investigators plan to retrospectively compare those two video sets regarding the polyp detection rate, withdrawal time and polyp identification characteristics of the AI system.
Interventions
DEVICEAI-assisted colonoscopy
Colonoscopies performed with assistance of an AI tool that highlights the areas that are susceptible to be a polyp.
Sponsors
Wuerzburg University Hospital
Study design
Allocation
NON_RANDOMIZED
Intervention model
PARALLEL
Primary purpose
DIAGNOSTIC
Masking
NONE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* Colonoscopies for Polyp detection
Exclusion criteria
* Colonoscopies for Inflammatory Bowel Disease (IBD). * Colonoscopies for work up of an active bleeding
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Polyp detection rate comparison | 45 minutes | Number of polyps detected divided by number of colonoscopies |
| Mean withdrawal time comparison | 45 minutes | Mean withdrawal time comparison |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| AI-Polyp bounding boxes - True Positive Evaluation | 45 minutes | 2 approaches: frame by frame analysis and temporal coherence analysis |
| AI-Polyp bounding boxes - False Positive Quantitative Evaluation | 45 minutes | 3 approaches depending on window-time detection |
| AI-Polyp bounding boxes - False Negative Evaluation | 45 minutes | Number of by bounding box missed polyps |
| Reaction Time Analysis | 45 minutes | Comparison time of polyp detection in a human vs machine approach |
Countries
Germany
Outcome results
None listed