A Multicentric Validation Study on the Accuracy of Artificial Intelligence Assisted System in Clinical Application of Digestive Endoscopy

Renmin Hospital of Wuhan University10,000 enrolled

Overview

This is an artificial intelligence-based optical artificial intelligence assisted system that can assist endoscopists in improving the quality of endoscopy.

Endoscopic diagnosis and treatment play an important role in the discovery and treatment of gastrointestinal diseases.With the rapid increase in the number of endoscopies, the workload of endoscopists increases further.The high workload reduces the quality of endoscopy, leading to incomplete coverage and incomplete detection of lesions.With the rapid increase in the number of endoscopies, the workload of endoscopists increases further.The high workload reduces the quality of endoscopy, leading to incomplete coverage and incomplete detection of lesions.Therefore, carrying out deep learning and other artificial intelligence methods has good academic research and practical value for improving the quality of endoscopic diagnosis and treatment.The research and development, testing and functional evaluation of artificial intelligence devices need to use a large number of endoscopic images, and at the same time, the effectiveness and safety of artificial intelligence devices need to be verified in different hospitals and environments.Based on this, our research group intends to collect endoscopic image data from different hospitals for training and validation of the model.

Study Type

OBSERVATIONAL

Enrollment

10,000

Conditions

Gastrointestinal Disease Endoscopy Artificial Intelligence

Interventions

Endoscopists refer to AI for diagnosisDIAGNOSTIC_TEST

The AI will provide a clinical diagnosis during endoscopy.

Eligibility

Sex: ALLMin age: 18 Years

Medical Language ↔ Plain English

Inclusion Criteria: 1. male or female aged 18 or above; 2. endoscopy and related examinations should be performed to further clarify the characteristics of digestive tract diseases; 3. be able to read, understand and sign the informed consent; 4. the researcher believes that the subject can understand the process of the clinical study, is willing and able to complete all the study procedures and follow-up visits, and cooperate with the study procedures; Exclusion Criteria: 1. have participated in other clinical trials, signed the informed consent and have been in the follow-up period of other clinical trials; 2. drug or alcohol abuse or psychological disorder in the last 5 years; 3. pregnant or nursing women; 4. subjects with previous history of gastrointestinal surgery; 5. the researcher considers that the subject is not suitable for endoscopy and related examination; 6. high-risk diseases or other special conditions that the investigator considers inappropriate for the subject to participate in the clinical trial.

Locations (1)

Renmin Hospital of Wuhan University

Wuhan, Hubei, China

Outcomes

Primary Outcomes

Accuracy

Calculate the accuracy of AI's judgment on images and videos. Accuracy is :

Time frame: 2020.1.12-2023.12.31

Sensitivity

Calculate the sensitivity of AI's judgment on images and videos. Sensitivity is : in the sample that is positive actually, the proportion that judges to be positive (for example, in the person that is really sick, be judged to be the proportion that is sick by the hospital), computation way is the ratio that true positive divides true positive add false negative (be positive actually, but judge is negative).

Time frame: 2020.1.12-2023.12.31

Specificity

Calculate the specificity of AI's judgment on images and videos. Specificity is : in the samples that are actually negative, the proportion of those that are judged negative (for example, the proportion of those who are not actually ill, who are judged by the hospital to be not ill) is calculated as the ratio of true negative divided by true negative + false positive (actually negative, but judged positive).

Time frame: 2020.1.12-2023.12.31

Positive Predictive Value (PPV)

The percentage of true positive people in positive test results indicates the probability that the positive test results belong to true cases.

Time frame: 2020.1.12-2023.12.31

Negative Predictive Value (NPV)

The percentage of true negative to negative test results indicates the probability that the negative test results are non-cases.

Time frame: 2020.1.12-2023.12.31

Receiver Operating Characteristic (ROC) Curve

Definition 1:The subject's operating characteristic curve is a coordinate graph composed of false positive rate as the horizontal axis and true positive rate as the vertical axis, and the curve drawn by the subject under specific stimulus conditions due to the different judgment criteria. Definition 2:ROC curves were created by plotting the proportion of true positive cases (sensitivity) against the proportion of false positive cases (1-specificity), by varying the predictive probability threshold.

Data from ClinicalTrials.gov

This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.

Secondary Outcomes

mean Average Precision (mAP)

mAP is setting a threshold for average precision and taking 1 or 0, and then taking the average of the sum of average precision divided by the number of values.

Time frame: 2020.1.12-2023.12.31

Sørensen-Dice coefficient (F1 score)

The Sørensen-Dice coefficient is a statistic used to guage the similarity of two samples. The F1 score is a weighted average of model accuracy and recall.

Time frame: 2020.1.12-2023.12.31

Recall Rate

The percentage of positive examples of predicted pairs in all samples of predicted pairs (including correct predicted positive examples and correct predicted negative examples).

Time frame: 2020.1.12-2023.12.31

Positive Likelihood Ratio

Time frame: 2020.1.12-2023.12.31

Negative Likelihood Ratio

Time frame: 2020.1.12-2023.12.31