ClearXview: an AI-powered orthopaedic imaging diagnostic solution
Lead Research Organisation:
IMPERIAL COLLEGE LONDON
Abstract
Knee osteoarthritis (KOA) is one of the most common causes of pain and disability, affecting over 367 million people globally. In the UK, KOA leads to more than £2.5 billion in annual healthcare costs. Detecting the disease early is essential to delay or avoid invasive treatments like joint replacement surgery. However, the current diagnostic process is often slow and uncertain, especially in the early stages of the disease.
A key issue is the communication loop between general practitioners (GPs) and radiologists. The way referrals are written can influence the wording of radiology reports, sometimes reinforcing assumptions rather than identifying disease with accuracy. At the same time, GPs rely on these reports to guide treatment decisions. This mutual dependence can introduce diagnostic bias and uncertainty, especially when early-stage signs of KOA are subtle. These challenges are made worse by a national shortage of experienced clinicians, contributing to delays in diagnosis and treatment.
The first imaging step for most patients is a 2D X-ray, which is quick and accessible but often lacks the detail needed to detect early joint degeneration. CT scans provide more accurate 3D information but are costly, expose patients to more radiation, and are not routinely used for KOA diagnosis. MRI offers excellent soft tissue detail but is expensive and not always available in the global healthcare system. This creates a gap between what standard screening imaging can provide and what clinicians need to make timely, accurate decisions.
This project aims to bridge that gap by validating a new artificial intelligence (AI) approach that reconstructs 3D images of the knee joint from just two standard X-rays. The system uses a generative AI model trained on real-world examples of matched 2D radiographs and 3D CT scans to produce detailed images that replicate CT-like bone structure.
Preliminary studies show that the model can reconstruct healthy anatomy with high accuracy. However, it is not yet known whether it can reliably capture disease-specific features such as joint space narrowing or osteophytes—key indicators used in KOA diagnosis. This project will generate the critical evidence needed to answer that question.
We will test the AI-generated 3D images using anonymised datasets of patients across different KOA severity levels. Clinical experts will assess the synthetic images alongside the real CT scans to judge how well the AI captures relevant disease features. This evaluation will provide essential data on the model’s accuracy and potential to support clinical decision-making earlier in the patient pathway.
If successful, this approach could improve the accuracy and speed of KOA diagnosis, reduce the need for advanced imaging, and support clinicians—particularly in primary and secondary care settings—with a more accessible diagnostic tool. By generating robust clinical evidence, this project will enable future development toward regulatory approval, NHS integration, and wider application across musculoskeletal care pathways.
A key issue is the communication loop between general practitioners (GPs) and radiologists. The way referrals are written can influence the wording of radiology reports, sometimes reinforcing assumptions rather than identifying disease with accuracy. At the same time, GPs rely on these reports to guide treatment decisions. This mutual dependence can introduce diagnostic bias and uncertainty, especially when early-stage signs of KOA are subtle. These challenges are made worse by a national shortage of experienced clinicians, contributing to delays in diagnosis and treatment.
The first imaging step for most patients is a 2D X-ray, which is quick and accessible but often lacks the detail needed to detect early joint degeneration. CT scans provide more accurate 3D information but are costly, expose patients to more radiation, and are not routinely used for KOA diagnosis. MRI offers excellent soft tissue detail but is expensive and not always available in the global healthcare system. This creates a gap between what standard screening imaging can provide and what clinicians need to make timely, accurate decisions.
This project aims to bridge that gap by validating a new artificial intelligence (AI) approach that reconstructs 3D images of the knee joint from just two standard X-rays. The system uses a generative AI model trained on real-world examples of matched 2D radiographs and 3D CT scans to produce detailed images that replicate CT-like bone structure.
Preliminary studies show that the model can reconstruct healthy anatomy with high accuracy. However, it is not yet known whether it can reliably capture disease-specific features such as joint space narrowing or osteophytes—key indicators used in KOA diagnosis. This project will generate the critical evidence needed to answer that question.
We will test the AI-generated 3D images using anonymised datasets of patients across different KOA severity levels. Clinical experts will assess the synthetic images alongside the real CT scans to judge how well the AI captures relevant disease features. This evaluation will provide essential data on the model’s accuracy and potential to support clinical decision-making earlier in the patient pathway.
If successful, this approach could improve the accuracy and speed of KOA diagnosis, reduce the need for advanced imaging, and support clinicians—particularly in primary and secondary care settings—with a more accessible diagnostic tool. By generating robust clinical evidence, this project will enable future development toward regulatory approval, NHS integration, and wider application across musculoskeletal care pathways.