Two-component Radiology-guided Autonomous Cascade Engine (TRACE)

Protocol for a Prospective Randomised Crossover Controlled Trial of the Artificial Intelligence-Assisted Decision-Making System for Gastric Cancer T-Staging (TRACE)

This study employed a prospective, randomised crossover trial design to evaluate the clinical utility of the TRACE artificial intelligence system for gastric cancer T-staging. A total of 54 radiologists from tertiary and non-tertiary hospitals, including both senior and junior practitioners, were enrolled. The study aimed to investigate whether AI-assisted diagnosis could improve the diagnostic accuracy of gastric cancer T-staging compared with independent interpretation by radiologists.

All participants were required to interpret 60 contrast-enhanced CT cases sequentially, completing two readings for each case: one without AI assistance and one with AI assistance; The order of the two readings was randomised, and a one-month washout period was observed between readings to eliminate memory bias. All cases were pathologically confirmed gastric cancer cases (stages T1-T4b), and the study simultaneously recorded the physicians' T-staging diagnostic results and the time taken per case. The 60 cases per radiologist were randomly selected from a pool of 1,000 histologically confirmed gastric cancer cases, stratified by pathological T stage T1-T4b. The reference standard was postoperative pathological T stage. The primary outcome was the change in T-staging accuracy between AI-assisted reading and standard (unaided) reading.The term "prospective" in this study refers to the prospective execution of radiologist enrollment, randomization, reading procedures, and data collection.

The TRACE trial is a prospective, randomized, crossover, controlled study evaluating an artificial intelligence (AI)-assisted decision system for T staging of gastric cancer based on CT images.

Background and rationale: Accurate preoperative T staging is critical for treatment planning in gastric cancer, but remains challenging due to reader variability and imaging limitations. The AI system was developed using deep learning with a large multi-center dataset to improve staging accuracy.

Study design: Eligible patients with pathologically confirmed gastric cancer will undergo preoperative contrast-enhanced CT. Each participant will be assessed twice in random order: once with AI assistance (AI arm) and once without (standard arm). A washout period will be applied between the two readings to minimize recall bias. Radiologists involved in the study are blinded to clinical and pathological reference standards.

Objective: To compare the T staging accuracy (primary outcome) between AI-assisted and standard reading, with secondary outcomes including inter-reader agreement, reading time, and diagnostic confidence.

Statistical methods: A crossover design will be used with a sample size calculated to detect a prespecified difference in overall accuracy. The primary analysis will employ a paired McNemar test or generalized estimating equation accounting for period and carryover effects. Subgroup analyses by tumor location, T category, and reader experience will be exploratory.

Data monitoring: No independent Data Monitoring Committee is required due to the low-risk nature of the diagnostic device. Adverse events related to the use of the software (e.g., workflow disruption) will be recorded and reported.

Ethics and dissemination: The protocol has been approved by the Ethics Committee of Liaoning Cancer Hospital & Institute. Written informed consent (online or paper-based) will be obtained from all participants. Results will be submitted for publication in peer-reviewed journals regardless of outcome.

Inclusion Criteria (Imaging Data)

1. Contrast-enhanced CT (CE-CT) images of gastric cancer patients from the Liaoning Cancer Hospital;
2. Patients with a definitive postoperative pathological diagnosis of gastric cancer and a clear T-stage classification (T1-T4, including T4a and T4b);
3. Imaging data must be complete and of sufficient quality to meet diagnostic and analytical requirements, with no significant artefacts or missing key data;
4. Complete clinical and pathological information must be available to establish a diagnostic gold standard for comparison.

Physician Inclusion Criteria (Image Readers)

1. Radiologists holding a valid medical licence;
2. From the radiology department of a Grade A tertiary hospital or a non-Grade A tertiary hospital;
3. Classified as senior or junior physicians based on clinical experience;
4. Voluntarily participating in this study and completing both the non-AI-assisted and AI-assisted image interpretation tasks.

Case Exclusion Criteria

1. Severe missing imaging data or quality failing to meet analysis requirements (e.g., severe motion artefacts);
2. Lack of clear postoperative pathological T-staging results;
3. Cases not involving gastric cancer or with incomplete pathological information;
4. Cases of duplicate enrolment or inconsistent data recording.

Physician Exclusion Criteria

1. Those unable to complete all image review tasks or demonstrating severe non-compliance;
2. Those who withdraw during the study period and are unable to provide complete data for both phases of image review;
3. Those who fail to complete the AI-assisted and non-AI-assisted interpretation processes as specified.

Withdrawal Criteria

1. Physicians who voluntarily withdraw from the study for personal reasons (e.g., time, health or work commitments);
2. Physicians who fail to complete the required image review tasks or have data missing in excess of the specified threshold;
3. Cases where critical data errors are identified during subsequent verification or where pathological results cannot be traced; Data found during the study to be non-compliant with ethical or quality control requirements must be excluded.