臨床研究等提出・公開システム

A Multicenter Single-Arm Prospective Performance Evaluation Study to Assess the Performance of an AI Instrument to Support Qualitative Diagnosis of Colorectal Polyps (MASTERPIECE study)

MASTERPIECE study (MASTERPIECE study)

176

A total of 176 subjects were enrolled and 139 subjects were included in the Full Analysis Set (FAS). In FAS, male were 95 subjects (68.3%) and female were 44 (31.7%). The median age was 70.0 years old (range:32 -85).

Of the 176 enrollment subjects, 9 were treated as enrollment violations because they had been undergone protocol endoscopy before enrollment, 165 were eligible subjects. Of the total eligible patients, 2 patients discontinued before endoscopy, then 165 patients were included in the safety analysis set. Of these, 26 had no lesions targeted for the study, then the full analysis set was 139 subjects. In FAS, 389 lesions were detected with endoscopy. Of these 389 lesions, 6 were not resected and 3 could not be diagnosed pathologically, then statistical analysis was performed for these 380 lesions (NEOPLASTIC : 300, HYPERPLASTIC (including normal mucosa and inflammation) : 80).

One subjects experienced Grade 2 Lower gastrointestinal hemorrhage.

The sensitivity, specificity and accuracy (95% confidence interval) for AI diagnosis with non-magnified BLI were 86.3% (81.9% - 90.0%), 70.0% (58.7% - 79.7%) and 82.9% (78.7% - 86.5%). For endoscopists diagnosis were 93.0% (89.5% - 95.6%), 70.0% (58.7% - 79.7%) and 88.2% (84.5% - 91.2%). In the analysis between the endoscopists' experience, for endoscopists with experiences less than 1500 subjects the sensitivity, specificity and accuracy (95% confidence interval) with non-magnified BLI were 91.2% (83.4% - 96.1%), 43.5% (23.2% - 65.5%) and 81.6% (73.2% - 88.2%), while for endoscopists with more than 1500 subjects were 93.8% (89.6% - 96.6%), 80.7% (68.1% - 90.0%) and 91.0% (86.9% - 94.1%).

The full analysis set consisted of 139 subjects with 380 lesions. Endoscopists diagnosed colorectal lesions with white light, non-magnified BLI, and magnified BLI, and endoscopic resection was performed after AI diagnosis. The sensitivity, specificity, and accuracy were evaluated using the histopathology results as the gold standard; the accuracy for AI diagnosis with magnified BLI was 88.7%, which is similar to previous reports.

Dec. 01, 2023

No

None

https://jrct.mhlw.go.jp/latest-detail/jRCTs072220003

Uraoka Toshio

Gunma University Hospital

3-39-15, Showa-machi, Maebashi, Gunma, 371-8511, Japan

uraoka@gunma-u.ac.jp

Sato Keigo

Gunma University Hospital

3-39-15, Showa-machi, Maebashi, Gunma, 371-8511 Japan

+81-27-220-8137

k.sato@gunma-u.ac.jp

Complete

April. 20, 2022

Interventional

single arm study

open(masking not used)

uncontrolled control

single assignment

diagnostic purpose

1.Patients between 20 and 85 years old at the time of registration who undergo colonoscopy

2.Patients who undergo colonoscopy (patients who meet any of the following conditions)

-Positive fecal occult blood

-Patients who need to undergo surveillance colonoscopy after polypectomy for colorectal polyp (including mucosal resection and submucosal dissection for early stage cancer)

-Patients who have gastrointestinal symptoms such as abdominal pain, diarrhea or constipation

-Patients who need to undergo screening colonoscopy for colorectal cancer

-Patients who are recommended to undergo colonoscopy by their physician

3.Patients who can understand the content forms, those who can make a decision by themselves and those who can give written consent

1.Patients with inflammatory bowel disease

2.Patients with familial adenomatous polyposis

3.Patients with a history of colorectal resection except appendicectomy

4.Patients with colorectal stenosis

5.Pregnant patients

6.Patients with abnormal blood coagulation function

7.Patients who cannot be managed with anticoagulant and/or antiplatelet agents according to the guidelines published by Japan Gastroenterological Endoscopy Society

8.Patients with severe organ failure

9.Patients who are not suitable for enrollment in the present study judged by their physician

Both

Patients who undergo colonoscopy

Using artificial intelligence (AI) for endoscopic diagnosis with colonoscopy

Patients who undergo colonoscopy

artificial intelligence (AI)

Accuracy of endoscopic diagnosis with AI

1.Sensitivity and specificity of endoscopic diagnosis with AI

2.Accuracy, sensitivity and specificity of endoscopic diagnosis made by endoscopists

3.Differences of accuracy, sensitivity and specificity of endoscopic diagnosis between trainees and experts of colonoscopy

4.Differences of accuracy, sensitivity and specificity of endoscopic diagnosis by characteristics of lesions (size, morphology and location of lesions) between endoscopists and AI

5.Differences of endoscopists confidence level on endoscopic diagnosis between diagnosing with AI and without AI

6.Rates of adverse events (perforation or postoperative bleeding, etc.)

+81-92-643-7171

Mar. 17, 2022

none