Evaluating a Text-Prompt AI Assistant for Chest CT Scans (AI-REPORT Study)

This study investigates whether an artificial intelligence (AI) system that drafts preliminary radiology reports can help experienced chest CT radiologists work faster while maintaining or improving report quality. The trial is conducted in two sequential phases. The first phase uses a set of complex, real-world historical cases. Radiologists interpret these cases both with and without the help of the AI-generated draft (AI-report) in a controlled, crossover study design. The second phase is a prospective, real-world deployment where the same AI-report system is integrated into the clinical workflow of participating radiologists as they interpret new, incoming chest CT scans in real time. We measure the time it takes to complete reports and, through blinded evaluations by other senior doctors, assess the quality of the final reports created with and without AI assistance. The goal is to determine if this AI tool can make radiologists' work more efficient and support high-quality patient care in actual practice.

• Detailed Description

1.1 Study Design

This is a two-phase, multicenter, multireader, multicase (MRMC) study designed to evaluate the real-world clinical utility of an AI report generation system (AI-report).

• Stage 1 (controlled crossover evaluation): This stage employs a retrospective, randomized, two-period crossover design. A curated set of complex historical chest CT cases, previously discussed in multidisciplinary team (MDT) meetings, is used. Each participating radiologist acts as their own control, interpreting the same cases both with and without the AI draft under controlled conditions.
• Stage 2 (prospective real-world deployment): This stage is a prospective, observational study. The validated AI-report system is deployed into the live clinical workflow of the participating radiologists. They use the system in real-time as they interpret new, consecutive chest CT scans from their clinical duties, allowing for evaluation in an authentic clinical environment.

1.2 Objectives

• Primary objectives: To evaluate the impact of the AI-report system on 1) radiologist efficiency (interpretation time) and 2) the clinical quality of finalized reports, assessed in both a controlled retrospective setting (Phase 1) and a prospective real-world setting (Phase 2).
• Secondary objectives: To assess the nature and clinical significance of edits made to AI drafts, and to evaluate system usability and integration into the routine reporting workflow.

1.3 Study Population

• Radiologist Readers: Board-certified radiologists with ≥ 3 years of independent thoracic imaging practice.
• Blinded Evaluators: Eleven senior clinicians from the original MDT panels that contributed the Phase 1 cases, responsible for blinded quality assessment.

1.4 Intervention

The intervention is the provision of a fully AI-generated draft radiology report (AI-report). In Phase 1, this is provided within a controlled reading platform for historical cases. In Phase 2, the system is integrated into the clinical Picture Archiving and Communication System (PACS)/Radiology Information System (RIS) to generate drafts for prospective, real-time cases.

• Study Procedures

Phase 1 (Retrospective Crossover): The 400 historical MDT cases are used. The study involves two reading rounds with a washout period. In each round, radiologists interpret a set of cases, with the AI condition (draft provided or not) randomized and crossed over between rounds. Interpretation time is recorded, and all finalized reports are collected for blinded pairwise comparison by the evaluator panel.

Phase 2 (Prospective Deployment): Following Phase 1, the AI-report system is activated in the clinical environment for participating radiologists. During a defined prospective observation period, the system generates drafts for eligible new chest CT scans. Radiologists use these drafts in their daily work. Reporting time and the AI drafts alongside the finalized human-edited reports are collected for analysis. Report quality in this phase is assessed longitudinally and through sampling.

• Outcome Measures

3.1 Primary Outcomes:

Efficiency: Change in median interpretation time per case with vs. without AI-report assistance (Phase 1) and the distribution of reporting times during real-world use (Phase 2).

Quality: Superiority score from blinded paired comparisons of AI-assisted vs. unassisted reports (Phase 1). Qualitative and quantitative assessment of report adequacy in the prospective cohort (Phase 2).

3.2 Secondary Outcomes:

Clinical significance of radiologist modifications to AI drafts (5-point scale).

System usability and workflow integration scores from post-study surveys.

• Statistical Analysis

Analysis will account for the MRMC design in Phase 1 using hierarchical models. Phase 2 data will be analyzed using descriptive statistics and statistical process control methods where appropriate. The two phases will be analyzed separately to provide insights into efficacy (Phase 1) and effectiveness (Phase 2).

Inclusion criteria

• Active board certification and ongoing routine clinical practice as an attending radiologist
• Independent institutional authority for chest CT image interpretation and final official diagnostic report issuance
• A minimum of three years of post-certification clinical experience in specialized thoracic imaging
• Legal and cognitive competence for study participation, with voluntary provision of written informed consent after full understanding of study purpose, procedures, risks and benefits

Exclusion criteria

• Direct participation in the development, training or validation of the trial's evaluated AI system
• Ongoing participation in concurrent studies with potential risks of interpretation bias, cognitive fatigue or study procedure interference (investigator-assessed)
• Any actual or perceived conflict of interest related to the evaluated AI system or its developers that may compromise objectivity in image interpretation and diagnostic reporting