Scalable Medical AI for Radiology Transformation – X-Ray (SMART-XR)

Background Chest X-rays (CXRs) play a pivotal role in identifying a broad range of thoracic conditions—from infections to lung cancer. However, NHS radiology services are under increasing strain due to rising imaging volumes and a shortage of radiologists—contributing to delays in reporting and prolonged patient wait times. Current UK regulations mandate that all CXRs be reviewed by a registered, appropriately trained clinician. Artificial Intelligence (AI) driven autonomous reporting has the potential to transform workflow efficiencies through reduced radiology workloads, ultimately reducing patient waiting times, improving time to diagnosis, and providing equitable access to imaging expertise across the healthcare system. Annalise Enterprise (AE) CXR is a validated, comprehensive AI clinical decision-support tool capable of detecting up to 124 findings. It has demonstrated high accuracy and efficiency in both large-scale validation studies and real-world NHS use. This project will rigorously evaluate the performance and safety of autonomous reporting using AE CXR, while capturing the perspectives of patients and clinical end-users to inform safe, acceptable, and effective implementation. Objectives To evaluate the safe and scalable implementation of autonomous reporting for CXRs using AE CXR, with the goal of substantially reducing radiologist workload and patient wait times across the NHS. We will move beyond current conservative paradigms of autonomous reporting by applying a clinically informed threshold that maximizes impact while ensuring patient safety. Methods A large-scale, retrospective evaluation of AE CXR for autonomous reporting will be performed using 12-months of consecutively acquired CXRs from two large NHS trusts. All AI findings will be categorised as either Remarkable or Unremarkable, based on vendor default criteria. The following will be undertaken: 1. Clinical thresholds and device configuration will be optimised to quantify the potential scale of autonomous reporting balanced by the impact of critical misses. 2. Quantify overall performance of AE CXR to distinguish between clinically significant abnormalities requiring radiologist review and minor findings suitable for autonomous reporting will be established. 3. Compare AI generated report formats and original radiology reports by clinical stakeholders (radiologists and referring clinicians) to determine the usefulness, accuracy, and clinical acceptability of each format. 4. Establish patient perspectives towards adoption of autonomous reporting in clinical workflows. 5. Develop a detailed economic model quantifying direct benefits (radiologist time, reporting turnaround, patient waiting time) and indirect system-level impacts of AE CXR implementation. Expected Results 1. Quantify the scale of opportunity for autonomous reporting and impact on patient waiting times 2. Evidence based configuration for safe and effective implementation of autonomous reporting 3. Preferred structure of AI-generated reports established 4. Practical guidelines for clinical integration that comprises patient communication frameworks to ensure trust and transparency 5. Economic impact of autonomous reporting at scale Conclusion This study aims to provide robust evidence for the adoption of autonomous AI reporting. Successful outcomes will transform radiology workflows, reduce patient wait lists, and improve timely patient care in the NHS trusts.