Machine Learning Driven Diagnosis of Low Bone Density on Plain-Film X-Rays

点击翻译按钮获取中文摘要

Osteoporosis is characterised by a degradation of the bone mass and an alteration of the bone microarchitecture. It is estimated that more than 200M people are suffering from osteoporosis worldwide and this number is expected to increase steadily due to changing demographics. If untreated, osteoporosis can lead to fractures which are not only disabling for the patient but cost care systems millions each year. A staggering 1 in 3 women over the age of 50 years and 1 in 5 men will experience osteoporotic fractures in their lifetime, therefore it is imperative that healthcare systems are poised to identify loss of bone mass as early as possible in order to reduce the incidence of fractures.One reason that fractures are so common is that ~75%(2) of osteoporosis patients remain undiagnosed (due to a lack of screening opportunities) and therefore miss the opportunity to receive treatment to lower their risk of fracture. This growing public health burden from osteoporosis and fragility fractures highlights the need to prioritise osteoporosis in healthcare systems immediately. As osteoporosis is highly preventable, a focus on effective management offers an opportunity for substantial savings in the future.In order to reduce the healthcare burden of osteoporosis and minimise the risks of fractures, patients need to be identified earlier so that they can receive effective treatment and interventions. Our proposed solution, OsteoSight, is a machine learning (AI) model that will take an X-ray image, analyse the bone structure within the image and label the bone as either 'osteoporosis' or 'no osteoporosis'. This approach means that any X-ray image taken in the secondary care setting can be leveraged as an additional opportunity to detect osteoporosis, reducing the onus on primary care providers to identify at-risk patients and refer them for diagnostic testing.Our innovation has the potential to increase osteoporosis case-finding by as much as 20%. This equates to a 5% improvement in patient outcomes (25,000+ fewer fragility fractures per year) and a 5% direct cost saving for the NHS (£220+ million per year). By leveraging existing image data, this approach requires no additional capital investments within the care settings and is non-invasive for the patients.