Colonic Diseases
Conditions
Keywords
Deep learning, Colonoscopy, Central Neural Networks
Brief summary
The purpose of this study is to develop and validate a deep learning algorithm to realize automatic recognition of colonic segments under conventional colonoscopy. Then, evaluate the accuracy this new artificial intelligence(AI) assisted recognition system in clinic practice.
Detailed description
Colonoscopy is recommended as a routine examination for colorectal cancer screening. Complete inspection of all colon segments is the basis of colonoscopy quality control, and furthermore improves the detection rates of small adenomas. Recently, deep learning algorithm based on central neural networks (CNN) has shown multiple potential in computer-aided detection and computer-aided diagnose of gastrointestinal lesions. However, there is still a blank in recognition of anatomic sites, which restricts the realization of AI-aided lesions detection and disease severity scoring. This study aim to train an algorithm to recognize key colonic segments, and testify the accuracy of each segments recognition as compared to endoscopic physicians.
Interventions
DEVICEAI assisted recognition of colonic segments
After receiving standard bowel preparation regimen, patients go through colonoscopy under the AI monitoring device. The whole withdrawal process is monitored by AI associated recognition system. Key colonic segments include ileocecal valve, ascending colon, transverse colon, descending colon, sigmoid colon and rectum. When typical anatomic sites are detected, the AI device will automatically captured relevant images and report the name of each segment on the screen. The operating endoscopy expert will give the final answer and judge the performance of AI, which is set as a golden standard. Then all the AI captured images will be reviewed by human group, which consists of three to five experienced endoscopic physicians.
Sponsors
Shandong University
Study design
Allocation
NA
Intervention model
SINGLE_GROUP
Primary purpose
HEALTH_SERVICES_RESEARCH
Masking
NONE
Eligibility
Sex/Gender
ALL
Age
18 Years to 70 Years
Healthy volunteers
No
Inclusion criteria
* Patients aged 18-70 years undergoing conventional colonoscopy
Exclusion criteria
* Known or suspected bowel obstruction, stricture or perforation * Compromised swallowing reflex or mental status * Severe chronic renal failure(creatinine clearance \< 30 ml/min) * Severe congestive heart failure (New York Heart Association class III or IV) * Uncontrolled hypertension (systolic blood pressure \> 170 mm Hg, diastolic blood pressure \> 100 mm Hg) * Dehydration * Disturbance of electrolytes * Pregnancy or lactation * Hemodynamically unstable * Unable to give informed consent
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| The accuracy of each colonic segment real-time recognition with deep learning algorithm. | 3 months. | The segmental recognition accuracy is the proportion of correctly recognized segments divided by the number of involved patients. The accuracy rate of ileocecal valve, ascending colon, transverse colon, descending colon, sigmoid colon and rectum will be separately calculated. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| The accuracy of total colonic segments recognition with deep learning algorithm as compared to endoscopic experts group. | 3 months. | The total recognition accuracy is the proportion of correctly recognized images divided by the number of AI captured images. Then all AI captured images will be reviewed by experts group to give a human evaluating rate. Two rates will be compared by student t test to analyze the difference. |
Contacts
Primary ContactXiuli Zuo, MD,PhD
Outcome results
None listed