Pulmonary nodules are frequently found during low-dose computed tomography (LDCT) health screening. The main challenge is not only detecting nodules, but also recommending the appropriate next step, such as routine follow-up, short-interval imaging follow-up, or specialist evaluation. This multicenter cluster-randomized trial will evaluate whether an artificial intelligence (AI)-assisted reporting workflow improves the appropriateness of pulmonary nodule management decisions in health examination settings without increasing under-management or missed referrals. Participating health examination branches, rather than individual participants, will be randomly assigned in a 1:1 ratio to a conventional reporting workflow or an AI-assisted reporting workflow. All final reports will be reviewed and signed by qualified physicians. An independent expert endpoint committee, blinded to study assignment and AI output, will determine the acceptable management range for each case. Approximately 2,000 adults with pulmonary nodules detected on LDCT will be included across at branches.
Age 18 years or older.
Undergoing chest low-dose computed tomography at a participating health examination branch during the study recruitment period.
At least one pulmonary nodule is identified on the index LDCT and requires risk stratification and a management decision regarding follow-up, repeat imaging, or specialist evaluation.
Thin-section reconstructed images are of sufficient quality for clinical interpretation and, where applicable, AI analysis.
Previously diagnosed lung cancer or currently receiving lung cancer-related treatment.
Known pulmonary metastasis from another malignancy.
Imaging findings that clearly require immediate entry into a lung cancer specialty diagnostic or treatment pathway and are not appropriate for routine pulmonary nodule risk-stratified management.
An urgent thoracic condition requiring immediate management, such as pneumothorax, large pleural effusion, or acute pulmonary embolism.
Parallel
AI-Assisted Pulmonary Nodule Reporting Workflow
Other interventionA locked-version artificial intelligence decision-support workflow applied after the physician records an initial assessment. The system displays pulmonary nodule location, size, density, morphologic features, estimated malignancy risk, and a suggested management category. The physician retains final responsibility. If the system fails, times out, or produces an abnormal output, the case returns to manual reporting. No online learning or automatic model updating is allowed during the trial.
Conventional Pulmonary Nodule Reporting Workflow
Other interventionStandard manual LDCT interpretation and pulmonary nodule management recommendation by qualified physicians using the participating branch's routine reporting process, without access to study AI output.
Proportion of Participants With an Appropriate Final Management Recommendation
Percentage of evaluable participants whose final physician recommendation is within the acceptable management range determined by the blinded independent expert endpoint committee. Recommendations are classified as routine or annual follow-up, short-interval imaging follow-up, or specialist evaluation/referral. The expert committee will complete adjudication using index data within approximately 30 days, but the participant-level outcome is the recommendation made at Day 0.
Proportion of Participants With Under-Management
Percentage of evaluable participants whose final physician recommendation is less intensive than the minimum acceptable management level determined by the blinded independent expert endpoint committee.
Proportion of Expert-Defined Referral Cases Missed by the Final Report
Among participants whom the blinded independent expert endpoint committee determines require specialist evaluation/referral, the percentage whose final physician report does not recommend specialist evaluation/referral. The denominator includes only expert-defined referral cases.
Proportion of Participants With Inappropriate Management Escalation
Percentage of evaluable participants whose final physician recommendation is more intensive than the maximum acceptable management level determined by the blinded independent expert endpoint committee.
Change in Decision Appropriateness After AI Review in the AI-Assisted Arm
In the AI-assisted arm, the physician's locked pre-AI decision will be compared with the final post-AI decision against the same expert-accepted management range. Results will include: (1) correction rate, the number changing from inappropriate to appropriate divided by the number with an inappropriate pre-AI decision; (2) AI-induced error rate, the number changing from appropriate to inappropriate divided by the number with an appropriate pre-AI decision; and (3) net correction rate, calculated as (number corrected minus number made incorrect) divided by all evaluable participants in the AI-assisted arm. These denominators must be locked in the statistical analysis plan before database lock.
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:
