This investigator-initiated, single-center study consists of a retrospective artificial intelligence model-development stage and a prospective physician reader-study stage. Deidentified chest CT examinations acquired during routine clinical care between January 1, 2021, and December 31, 2024, will be used to develop, validate, and lock artificial intelligence models for lung cancer-related imaging tasks.
In the prospective stage, approximately 12 to 15 physicians with experience in chest CT interpretation will complete two reading sessions in randomized order: unaided interpretation and AI-assisted interpretation. The sessions will be separated by a washout period of at least 4 weeks. The primary objective is to compare diagnostic performance between AI-assisted and unaided interpretation. Secondary objectives include reading time, diagnostic confidence, inter-reader agreement, and errors related to incorrect AI suggestions.
All readings will be performed in an offline research environment. AI outputs will not be used for patient care, and the study will not add imaging examinations, treatment, or follow-up for patients.
Physicians with education or training relevant to medical imaging
Experience in interpreting chest CT examinations
Ability to complete the required study training and both reading sessions
Willingness to provide written informed consent and comply with study procedures
Failure to complete the required study training
Inability or unwillingness to complete both reading sessions as required
Any protocol deviation likely to compromise the validity of the reader-study data
Crossover
Chest CT Artificial Intelligence Decision-Support System
DeviceA locked research-use-only artificial intelligence system analyzes deidentified chest CT images and provides decision-support outputs to participating physicians. The system is evaluated only in an offline reader-study environment and is not used for actual patient care.
Difference in Diagnostic Accuracy Between AI-Assisted and Unaided Chest CT Interpretation
Diagnostic accuracy will be calculated as the proportion of case-level interpretations that agree with the prespecified reference standard based on pathology or clinical follow-up. The paired difference in diagnostic accuracy between AI-assisted and unaided interpretation will be estimated across participating physicians and cases using a multi-reader multi-case analysis and reported with a 95% confidence interval. Higher accuracy indicates better diagnostic performance.
Click on the lead sponsor to view all of their trials.
This trial is not recruiting at the moment. You can still explore other options:
