Neural networks could improve patient outcomes and reduce healthcare costs

A new machine learning process designed to identify and classify hip fractures has been shown to outperform human clinicians.

Two convolutional neural networks (CNNs) developed at the University of Bath were able to identify and classify hip fractures from X-rays with 19% higher accuracy and confidence than hospital clinicians, in the results published this week in Nature Science Reports.

The research team, from Bath’s Center for Therapeutic Innovation and Institute for Mathematical Innovation, along with colleagues from Royal United Hospitals Trust Bath, North Bristol NHS Trust and Bristol Medical School, set out to create the new process to help clinicians to fracture the hip. more effective care and to support better patient outcomes.

They used a total of 3,659 hip X-rays, graded by at least two experts, to train and test the neural networks, which achieved an overall accuracy of 92% and 19% better accuracy than hospital clinicians.

Efficient processing is crucial to managing high costs

Hip fractures are a major cause of morbidity and mortality in the elderly, resulting in high health and social care costs. Staging a fracture before surgery is essential to help surgeons select the right interventions to treat the fracture, restore mobility, and improve patient outcomes.

The ability to quickly, accurately and reliably classify a fracture is essential: delays in surgery of more than 48 hours can increase the risk of adverse effects and mortality.

Fractures are divided into three classes – intracapsular, trochanteric or subtrochanteric – depending on the part of the joint in which they occur. Some treatments, which are determined by the classification of the fracture, can cost up to 4.5 times more than others.

In 2019, 67,671 hip fractures were reported to the UK National Hip Fracture Database, and given projections of population aging over the coming decades, the number of hip fractures is expected to increase worldwide, especially in Asia. Worldwide, around 1.6 million hip fractures occur each year with a substantial economic burden – around $6 billion a year in the US and around £2 billion in the UK.

The longer-term outcomes for patients are just as important: people who suffer a hip fracture have twice the age-specific mortality in the following year than the general population. Thus, according to the team, the development of strategies to improve the management of hip fractures and their impact on morbidity, mortality and the costs of healthcare delivery is a high priority.

Growing demand in radiology departments

A critical issue affecting the use of diagnostic imaging is the mismatch between demand and resources: for example, in the UK the number of x-rays (including x-rays) performed each year has increased by 25% between 1996 and 2014. Increasing demand in radiology departments often means they cannot report results in a timely manner.

Professor Richie Gill, lead author of the paper and co-director of the Center for Therapeutic Innovation, says: “Machine learning methods and neural networks offer a new and powerful approach to automating diagnostics and predicting outcomes, so this new technique we have shared Although the classification of fractures so strongly determines surgical treatment and therefore patient outcomes, there is currently no standardized process for determining who determines this classification in the UK – let it be done by orthopedic surgeons or radiologists specializing in musculoskeletal disorders.

“The process we have developed could help standardize this process, achieve greater accuracy, speed up diagnosis and alleviate the bottleneck of 300,000 x-rays that remain unreported in the UK for over 30 days. .”

Dr Otto Von Arx, consultant in Orthopedic Spine Surgery at Royal United Hospitals Bath NHS Trust, and one of the paper’s co-authors, adds: “‘As trauma clinicians, we constantly strive to deliver excellence in care to our patients and the healthcare community. through accurate diagnosis and cost-effective medicine.

“This excellent study has provided us with an additional tool to refine our diagnostic arsenal in order to provide the best care to our patients. This study demonstrates the excellent value of the collaboration between the RUH and the lead researcher, the University of Bath.

The study was funded by Arthroplasty for Arthritis Charity. The NVIDIA Corporation provided the Titan X GPU that performed the machine learning, through its academic grant program.

Sherry J. Basler