AI-Assisted Chest X-Ray Interpretation in Resource-Limited Settings: LuAna Stepped-Wedge Trial Protocol (Preprint)
Bibliographic record
Abstract
Background: Artificial intelligence (AI) has the potential to transform chest radiography interpretation by enhancing diagnostic accuracy, identifying subtle findings, reducing errors, and helping prioritize patient care. Although chest radiography remains a cost-effective and widely used imaging tool, its effectiveness is limited by overlapping anatomy and variability in clinical expertise. Integrating AI can help overcome some of these challenges, especially in resource-constrained settings. However, robust validation in real-world clinical contexts is essential before widespread implementation. This study protocol evaluates whether AI assistance improves general practitioners' ability to detect radiographic findings on chest radiography in adults with respiratory complaints or those undergoing treatment for respiratory diseases compared with unaided interpretation. Potential benefits include increased diagnostic safety, higher physician confidence, more efficient workflows, and expanded access to expert support in underserved areas. Objective: This study aims to evaluate whether AI assistance enhances physicians' ability to detect key radiographic abnormalities, including consolidation or pulmonary opacity, pneumothorax, atelectasis, pleural effusion, and cardiomegaly. The primary outcome is the difference in physicians' diagnostic accuracy (per examination) when assisted by the AI tool compared with usual practice, using expert radiologist consensus as the reference value. Methods: This study is a protocol for a multicenter, stepped-wedge, cluster-randomized clinical trial following the CONSORT-AI (Consolidated Standards of Reporting Trials-Artificial Intelligence) extension and SPIRIT-AI (Standard Protocol Items: Recommendations for Interventional Trials-Artificial Intelligence) guidelines. The intervention involves the diagnostic support solution for chest radiography, Lung Analysis (LuAna), an AI-powered chest X-ray interpretation tool developed in partnership with the Brazilian Ministry of Health. Across 9 cities in Brazil, clusters will transition monthly from unaided chest X-ray interpretation by general practitioners to AI-assisted interpretation, with performance benchmarked against thoracic radiologists. The stepped-wedge design ensures that all clusters receive the intervention, reflecting real-world coordination, enhancing acceptability, improving statistical power, and strengthening causal inference through repeated measures. Diagnostic performance will be compared with a reference standard established by thoracic radiologists. Results: This project was funded in October 2024 (following ethics approval by the institutional review board). Data collection commenced in January 2026 and is projected to be completed by September 2026, marking the end of the trial period. As of November 2025, 3 centers were fully prepared for enrollment initiation. The LuAna clinical trial is currently ongoing, with data analysis (including statistical analyses) forecasted to be finalized by November 2026. Results are expected to be published by January 2027. Conclusions: This intervention is expected to enhance clinical decision-making by supporting earlier treatment initiation and more appropriate diagnostic pathways for patients with respiratory symptoms while maintaining a favorable safety profile and high physician usability. Findings from this trial will provide real-world evidence on the clinical utility of AI-assisted chest radiography. If effective, LuAna may leverage its scalability and equity advantages to become a replicable model for integrating AI into routine imaging workflows worldwide, especially in regions with limited access to specialist care.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.030 | 0.078 |
| Meta-epidemiology (narrow) | 0.003 | 0.003 |
| Meta-epidemiology (broad) | 0.004 | 0.003 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.004 | 0.003 |
| Scholarly communication | 0.004 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.006 | 0.007 |
| Insufficient payload (model declined to judge) | 0.104 | 0.024 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".