HealthVCF for detecting vertebral compression fractures on CT scans

Level of innovation

Two experts felt that using artificial intelligence (AI) to identify vertebral fractures is novel. One considered the technology to be a minor variation on standard care because they felt that radiologists are more than capable of identifying vertebral fractures on CT scans without using the technology. All experts noted that these types of fractures are often missed, overlooked, or not reported by radiologists. One of the reasons given for this was that vertebral fractures are often not linked to the main clinical issues the reporting radiologists have been asked to consider. Two of the experts identified an alternative competing AI technology. The remaining expert was not aware of any specific competing products but noted that the number of AI technologies in radiology is increasing and it is likely that other products will provide a similar function.

Potential patient impact

Earlier detection of incidental fractures was highlighted by experts as a benefit of the technology, leading to earlier diagnosis of osteoporosis or reduced bone mineral density, more timely clinical management, and reduced risk of future fractures. One expert said all patients having CT scans may benefit because some of the common clinical reasons for having a CT scan (such as malignancies, liver diseases, chronic cardiac and respiratory conditions or their treatments) are also risk factors for osteoporosis. Another said that patients over 55 having routine CT examinations of the chest, abdomen, and pelvis would benefit most from the technology. The remaining expert also agreed that older people are more likely to benefit from the technology than younger people because of the higher prevalence of vertebral fractures.

Potential system impact

Increased identification of patients with osteoporosis and reduced numbers of future fractures was highlighted by experts as being a potential benefit to the healthcare system. One expert noted that spinal and hip fractures cost the NHS more than £5 million and that the technology could help reduce GP and hospital visits and the numbers of acute hip operations needed.

One expert felt that although adopting the technology would incur a small initial investment, this is likely to be offset in the long-term with savings from fewer primary care and hospital visits, and a decrease in the number of hospital admissions and surgical episodes. One said that adopting the technology would increase costs in the short term and that the cost benefits from reduced future hip fractures would not be seen for several years. One of the experts said that the technology could be cost saving compared with standard care.

Experts highlighted that using the technology would involve more radiologist time to review the reports and it would also increase the referral rates to bone health teams. One noted that an increased need for radiologist time could affect waiting times for reports to be issued. One noted that fracture liaison services would need additional resources to manage increased patient numbers and there would be an increased cost associated with treating patients with osteoporosis.

One expert said that minimal radiologist training (10 to 20 minutes) would be needed, as well as local PACS support training. Another agreed that minimal training would be needed but noted the potential need for radiology information systems integration update training. The expert also noted that the on-premises service needs a small computer server so an area in server rooms with intranet and power connections may be needed.

General comments

The experts said that the technology is not widely used across the NHS. Only 1 NHS trust was identified by experts as using the technology in clinical practice. Data security, availability of IT resources and limits on trust expenditure were identified as potential issues that could prevent adoption. One expert stated that robust quality assurance and accompanying governance structure will need to be in place before adopting these types of technology. One expert noted that the performance of the technology (sensitivity of 59%) appears lower than would be expected for an AI trained to recognise only 1 pathology. They also said that understanding how the AI is predicting future fracture risk from the CT data is important and that this is currently unclear.