.jpg)
Audio Version: Press Play to Listen
From fracture detection to bone density analysis – how AI is enhancing the diagnosis of a host of bone-related disorders.
Pertaining to the diagnosis of disorders relating to bone structures, joints and their associated soft tissues, musculoskeletal (MSK) radiology was one of the earliest subspecialties of radiology. From diagnosing fractures and breaks to chronic diseases including arthritis and osteoporosis, the field of MSK radiology spans a host of potential conditions across the entire human body.
The aging population is driving significant increases in MSK radiology workload. People are living longer and remaining active for longer than previous generations, which is resulting in an increase in overuse injuries, as well age-related diseases like osteoarthritis. From 1990 to 2020, global cases of MSK disorders rose by more than 120% and are projected to increase by another 115% by 2050.
Couple this with the time it takes to interpret and report on cases and it’s no surprise to learn that the burden being placed on MSK radiologists is growing, with increasing workload leading to burnout. Indeed, Society of Skeletal Radiology members reported significantly higher and more severe levels of burnout (80.5%) compared to other radiologists and physicians.
As in all areas of radiology, missed findings are a key issue in MSK – delays in diagnosis can negatively influence a patient’s outcomes and cause unnecessary pain and suffering. Fracture detection and interpretation is one of the most commonly missed diagnoses in radiology, accounting for the majority of errors in emergency departments and the majority of malpractice claims made against radiologists.
Another challenging area within MSK is around inter-reader reproducibility – particularly when it comes to quantification of an observed condition, such as osteoarthritis.
All these factors mean that MSK radiologists are under mounting pressure to maintain efficiency amidst escalating imaging volumes. And, with the demand for specialized radiologists outstripping supply, the industry is increasingly looking to the power of AI to help address these challenges. These days, rather than asking if AI will supplant MSK radiologists, the question most are asking is how can it be harnessed to augment their expertise?
Let’s look at some of the key areas where AI is already playing a significant role in MSK radiology…
Fracture detection and triage
Given the stats on missed fractures, it’s perhaps no surprise that one of the first areas of MSK that AI was developed for was bone fracture detection on X-rays. Despite the growing use of modalities like CT and MR, X-rays are still the most common form of imaging in the MSK field, and, in many ways, are the bread and butter for the specialty.
From making the binary decision of “fracture” or “no fracture”, to identifying the areas where fractures are found AI is already showing promising results. Fracture detection algorithms are helping improve diagnostic confidence and avoid missed findings, increasing throughput by quickly identifying fractures, routing complex cases to specialist MSK clinicians, and even assisting in reporting.
The regulatory framework that exists in the US is also favorable for AI solutions in fracture detection, and there are several solutions for hospitals to consider:
AZmed’s Rayvolve is a computer-aided diagnostic tool designed to detect fractures in X-rays. A clinical study conducted at University Hospitals in Cleveland showed the tool reduced reading time by 27%, with 99% sensitivity and 89% specificity.
“We also work with an outpatient network of 250 centers all across the US, and their main challenge was turnaround time - making sure they can send the report to the patient as quickly as possible,” says Bastien Manceau, Head of Sales at AZmed. “Using Rayvolve, they were able to reduce a 48-hour turnaround time to just 8.5 hours, because we flagged the positive cases for them to report first.”
Milvue, a European company, enhances emergency department workflows with its innovative Milvue Suite AI solution. Their TechCare Alert module specializes in the detection and triage of fractures, joint effusions, dislocations, and chest diseases in radiographs, streamlining the diagnostic process. The technology demonstrated a greater than 90% success rate when tasked with assessing a subgroup of challenging cases, originally misdiagnosed in the emergency department.
Aissa Khelifa, Chief Executive Officer at Milvue, asserts that the widespread adoption and effective implementation of AI in healthcare are crucial. He believes that demonstrating such clear benefits to hospitals, physicians, and patients is essential for the democratization of AI technology.
BoneView, the first clinical AI application from Gleamer, efficiently identifies fractures, effusions, dislocations, and bone lesions. In clinical studies, the tool demonstrated a 30% reduction in missed fractures, as well as up to a 36% reduction in reading time.
Finally, trained on more than 300,000 radiographs, RBfracture from Radiobotics detects fractures in X-rays across the entire body. A study conducted at a UK hospital showed using the tool delivered a 62% reduction in missed fracture rate compared to the same period the previous year.
In addition to the diagnostic confidence and time-saving benefits of using AI tools like these for fracture detection, healthcare providers may also be able to use them to streamline their workflows. There may be many different reasons that a patient may be assessed for fracture, and hospitals may have many different workflows, depending on the point of entry. An additional benefit of using AI is that it may provide the opportunity to consolidate multiple workflows into a single pathway.
Osteoarthritis grading
Driven by the aging population, arthritis is a growing health issue, with one in five US adults estimated to have some form of the disease. Osteoarthritis (OA) is the most common form of arthritis, thought to affect more than 32.5 million US adults, and OA of the knee is said to be responsible for more than $27 billion in healthcare costs annually.
X-rays have long been the standard imaging option for knee OA diagnosis, and the volumes of radiographs to be assessed has steadily increased along with the growing prevalence of the condition. In addition, measurement of severity of the condition can differ between radiologists. The use of AI can help accelerate the diagnostic process and provide standard measures of severity.
KOALA from ImageBiopsy Lab automatically detects signs of knee OA based on standard joint parameters and criteria from X-rays of the knee. A clinical study showed that the tool improved agreement rates between readers by 39% for knee OA diagnosis.
“The big challenges in medicine are always around medical errors and patient safety – you don't want to miss anything, and you want to have standardized output and reports – that’s what we deliver,” says Marcel Cimander, Channel Partner Manager at ImageBiopsy Lab. “And, of course, it's also always important to be efficient. With our fully automated AI pre-analysis, you don't have to do all the measurements or annotations yourself, which can be quite tedious, and the reimbursement also is not very high for on conventional X-ray images.”
Rated as “good-to-excellent” compared to human experts, RBknee from Radiobotics provides knee OA gradings at an MSK expert level and drafts a template report to enable faster turnaround times. “The tool is currently being evaluated as an unbiased first reader at the Copenhagen University Hospital, which is also exploring its potential to provide a unified pathway for patients with knee osteoarthritis and reduce the need for unnecessary MRI scans.” says Patrick Hynes-Foy, Head of Sales at Radiobotics.
Automating MSK measurement
Measurement is a significant factor throughout the MSK field. Whether for the assessment of scoliosis, hip dysplasia, lumbar spine, or leg angle and length, radiologists are frequently required to take measurements from various radiology modalities to support the diagnosis and evaluate the severity of a wide range of conditions.
In addition to being time consuming to conduct, these measurements are subjective, which can lead to variance in results depending on the reader. Increasingly AI tools are emerging that can automate and standardize measurements, accelerating the process and ensuring reproducibility of results.
With the goal of reducing measurement time and missed findings, CoLumbo is an AI tool that analyzes MRI images of the lumbar spine to detect abnormalities and provide information on suspected pathologies. An internal clinical evaluation report found that radiologists spent 48% less time to read patient scans with CoLumbo compared to reading without CoLumbo.
Providing fully automated measurements for the feet, legs, pelvis, hips, and spine, BoneMetrics from Gleamer aims to standardize measurements on X-rays. Shown in clinical studies to match the expertise of MSK radiologists, the tool aims to save time, minimize variability, and ensure reproducibility.
“BoneMetrics is a game-changing solution that streamlines the process of measuring bones efficiently and accurately, allowing our center to save time, improve workflow, and make a better report for clinicians,” says Dr Aurelien Lambert, a radiologist at IM2P in France.
When it comes to the assessment of hip degeneration, HIPPO from ImageBiopsy Lab aims to help radiologists improve the efficient diagnosis of femoroacetabular impingement and hip dysplasia. The tool provides objective and standardized measurements of the most important hip angles from X-rays, reducing reading and reporting time by 80% and eliminating subjectivity, as found a US Market Survey ImageBiopsy Lab conducted.
Finally, Milvue’s TechCare Bones and Spine AI modules automate over 30 MSK measurements across various body parts including hands, shoulders, spine, pelvis, hips, legs, and feet. These modules are designed to enhance radiologist productivity by accelerating measurement times and to support detailed orthopedic assessments through the automation of critical MSK imaging measurements.
Incidental osteoporosis detection
Another condition on the rise due to the increasingly aging global population is the bone disease osteoporosis. Characterized by low bone density, the condition is thought to affect up to half of Americans over the age of 50, although many remain undiagnosed.
Vertebral compression fractures are often a sign of osteoporosis and yet only a quarter of vertebral fractures are correctly and accurately reported as an incidental finding. As a result, an opportunity exists to identify the condition from CT scans conducted for other conditions, and AI companies have been quick to respond.
HeartLung.AI has developed, and received FDA clearance, for their opportunistic osteoporosis screening AI tool called AutoBMD™ that analyzes CT scans obtained for other purposes and generates a DEXA-equivalent bone mineral density (BMD) report with Z-score and T-score which is reimbursed by Medicare and private insurance carriers under CPT 77078. The AutoBMD AI eliminates the need for patients to undergo another X-ray scan for bone density measurement. Therefore, it saves patients extra radiation, and saves the healthcare system extra cost.
“More than 20 million people get a chest or abdominal CT scan every year, and none of those people are currently alerted if they have osteoporosis and what their BMD score is,” says Morteza Naghavi, M.D., Founder of HeartLung.AI. “AutoBMD is currently the only FDA-cleared AI tool that can both save radiologists time and simultaneously reduce the need for patients to have to get a DEXA scan to get the same report.”
HealthOST Bone solution, an AI tool developed by Nanox AI, provides qualitative and quantitative analysis of the spine from CT scans to support clinicians in the evaluation and assessment of musculoskeletal disease, such as osteoporosis. Early findings from a clinical study in the UK revealed that the tool demonstrated a six-fold improvement in the detection of spine fractures, an early sign of osteoporosis.
“Our algorithm automatically detects vertebral compression fractures and low bone mineral density to help identify patients at risk of osteoporosis, without the need for discrete imaging,” says Dr. Orit Wimpfheimer, Chief Medical Officer at, Nanox. “HealthOST has the option to automatically populate the results in the radiology report, which can then be used to help physicians make clinical care recommendations and increase early diagnosis and treatment for patients at high risk.”
Finally, providing accurate detection and labeling of vertebral fractures, ImageBiopsy Lab’s FLAMINGO tool detects vertebral fractures as secondary radiological findings on CT scans conducted for other conditions. The tool has demonstrated greater than 90% accuracy when it comes to detecting patients with at least one vertebral fracture.
But how to choose?
In the field of MSK radiology, the adoption of AI clearly offers hospitals and imaging centers a potential pathway towards enhanced efficiency and streamlined operations. However, as this article has already shown, there are a wide range of options to consider, and one of the challenges facing healthcare providers can be in determining which AI solution is the best fit for their specific needs.
This challenge, plus many others, can be effectively addressed by partnering with a single strategic AI platform provider, like Blackford, which allows organizations to test AI solutions using their own data before making their purchase decision.
“The growth of AI tools for MSK radiology, and in many other areas as well, has led to the emergence of a crowded and complex market,” says Ben Panter, CEO of Blackford. “We understand the challenges that healthcare providers face in choosing the right solutions and that certain applications might work well in one institution but not in another. “By having an extensive portfolio of applications covering important use case areas, such as fracture detection, we help healthcare providers evaluate which application works best for their unique needs.”
In general, centralizing access to AI solutions through a dedicated platform empowers healthcare providers to efficiently evaluate, implement, and maintain a range of imaging AI applications, thus maximizing the use of available resources. This approach not only reduces deployment timelines and financial overheads but also facilitates training, monitoring, and support efforts by focusing on a unified system.
Additional benefits of a platform-centric approach include simplified processes, consolidating various aspects such as contact management, deployment procedures, and ongoing support into a single interface. Crucially, a platform integrates with existing hospital systems, ensuring smooth performance, especially in handling DICOM interfaces.
Moreover, platforms alleviate the burden of analysis paralysis by offering access to a portfolio of pre-vetted AI applications spanning various medical disciplines beyond MSK, including cardiology, neurology, pulmonology, trauma, and women’s health. Hospitals can procure comprehensive bundles of AI applications through a single contract, eliminating the complexities associated with acquiring individual algorithms.
Ultimately, by embracing a single-platform strategy, clinicians and their institutions can gain access to a diverse array of AI solutions, enabling swift adoption of new technologies and facilitating the evaluation of applications tailored to their specific needs.