Macadamian, Radiobotics, and Bispebjerg Hospital Partner on AI Solution for Radiology: Interview

While the number of clinical data points available per patient continues to increase exponentially, the number of providers and specialists available to interpret that data fails to keep pace. As a result, technology-driven automation is becoming more important to quickly assess and triage patients as new information becomes available.

One clinical area where this disparity between available data and providers exists is radiology. Fast, accurate diagnosis is not always available, particularly in remote regions where specialists are not immediately accessible. To address this challenge, software design and development firm Macadamian is teaming up with both Denmark-based Radiobotics, an AI startup, and Bispebjerg Hospital to create X-AID. Initially aimed at analyzing X-ray images from non-acute musculoskeletal patients, X-AID is a clinical decision support solution incorporating both machine learning (ML) and deep convoluted neural networks (CNNs).

Earlier this year, the trio of partners secured $495k in Eurostars Eureka innovation project funding and was ranked as the number one project among a field of 283 eligible applicants. To learn more about the project, Medgadget heard from Timon LeDain, Director of Emerging Technologies at Macadamian.

Medgadget: What challenges is the X-AID solution seeking to solve and how does it solve them?

Timon LeDain: X-AID is focused on addressing a number of challenges including the exponential increase in requirements for imaging data resulting in radiologists being overloaded and unable to deliver radiological reports in a timely manner. Increases in radiologists’ workloads lead to increasingly rushed diagnosis which in turn leads to errors or mistakes in diagnostic reports. With longer wait times for imaging examinations and diagnoses, early detection, diagnosis, and treatment are impacted. Today, subjective interpretation and analysis of medical images has led to a large degree of inter-observer variability.

To address these challenges, Bispebjerg Hospital, Radiobotics, and Macadamian Technologies partnered to design and develop a machine learning-based clinical decision software platform that automatically analyses x-rays of the musculoskeletal system and generates the medical reports for review by radiologists.

The algorithms used were trained with data screened and validated by the clinical radiological partners to mimic expert accuracy in imaging diagnosis. Once an x-ray image enters into the system, it is immediately segmented and then the software calculates the angles and distances between bony structures and locates clinically relevant and critical areas associated with joint disease or bone fractures all identified in a standardized manner.

An objective data-based report is accessible immediately. Hospital information and communications technology systems are complex, and legacy PACS/RIS systems (picture archiving and communication system/radiological information system) are closed ecosystems that are difficult to integrate into. X-AID was architected to integrate with a range of provider cross-platform applications in addition to ensuring that it meets regulatory compliance requirements for medical devices and data security and privacy.

Medgadget: What differentiates X-AID from other image analysis solutions leveraging AI and/or ML?

LeDain: The benefits of the X-AID solution begin with the origins of our training data. Denmark has been digitizing x-rays for the past 20 years which means that through this collaboration, we have access to a rich set of training data combined with very accurate reports from some of the best doctors.

The problem that the team is tackling is fairly unique in that we are addressing the high volume of routine x-rays for musculoskeletal conditions that are exacerbated by a lack of radiologists around the world. This is in contrast to other solutions tackling complex problems with lower image volumes. As a result, the X-AID solution distinguishes itself in terms of its emphasis on workflow management to drive hospital efficiencies and reduced wait times for x-ray results.

Medgadget: Regarding the collaboration between Macadamian, Radiobotics, and Bispebjerg Hospital, can you share how these three partners came together and the role each plays in developing and launching X-AID?

LeDain: The collaborative partnership came about through a global grant-funded consortium between the Danish Bispebjerg Hospital’s Radiology Department, Radiobotics, and Macadamian Technologies. The consortium received grant funding from the Eurostars Eureka Project Award for Innovation in addition to a grant from The Canadian Government’s Industrial Research Assistance Program (IRAP) designed to accelerate the research and development projects of Canadian innovators.

Radiobotics developed and obtained the CE Mark for an AI algorithm called RB Knee that detects osteoarthritis in knee joints. This is just one of the algorithms they are working on integrating into the X-AID platform.

Dr. Mikeal Boeson, Head of Radiology at the Bispebjerg Hospital in Copenhagen, Denmark is the Clinical Researcher who brings clinical and domain expertise. Dr. Boeson is overseeing the clinical validation trials of the X-AID solution.

Medgadget: More specifically, tell us about Macadamian’s HealthConnect platform and the role it plays in the X-AID solution.

LeDain: Macadamian HealthConnect is a digital platform as a service that our software development team uses to streamline the development of cloud-connected healthcare applications, like X-AID, by standardizing the core building blocks to ensure that data privacy, compliance, and interoperability requirements are met from the get-go.

Macadamian HealthConnect has been deployed successfully to gather patient-reported outcomes for Parkinson’s patient studies and to securely share MRIs from patients with rare forms of muscular dystrophy in order to develop a diagnosis through a network of global collaborators.

For X-AID specifically, the platform will be used to streamline the entire AI development and deployment lifecycle, beginning with supporting training data generation by allowing experts around the world to annotate images for training purposes after they have been anonymized and securely uploaded to our cloud platform. HealthConnect also supports various hospital IT system integrations that will support the clinical validation and deployment phases through pre-built PACS, RIS, and EMR connectors. Finally, the platform enables data to be gathered during clinical trials and even post commercialization phase that will allow the AI developers to continue to monitor and improve their algorithms over time.

Medgadget: Is X-AID already being used in a clinical setting? If so, what are some of the outcomes realized thus far?

LeDain: The X-AID solution is already being evaluated by radiologists at Bispebjerg Hospital. At RSNA this week, we reported benefits realized in early testing including:

High agreement between AI delivered reports and those from radiologists

Higher agreement with expert radiologists highlighting the potential for support and training of junior radiologists

Increased consistency representing a significant benefit of an algorithm which will always return the same findings for the same image