Insights

Integrating AI into Aging Medical Imaging Systems in a Cost-Constrained Era

The integration of artificial intelligence (AI) in medical imaging has been a rapidly evolving and promising area of development. A quick look at this year’s RSNA program will attest to the overarching interest in AI, and for good reason. AI technologies, particularly machine learning algorithms, have shown great results within the medical diagnostic imaging industry, especially in improved diagnosis and detection through AI assisted image enhancement and analysis.

 

The explosion of AI capability and the product implications are only a part of the pressures on imaging companies. The medical imaging industry has been under significant cost pressures for more than a decade. That pressure to reduce costs is juxtaposed with the usual pressures to improve the performance of the imaging systems. Adding AI assisted image processing to legacy systems can significantly improve the quality of the resultant images, without expensive redesigns and increased hardware costs of new image acquisition systems.

 

AI can have a very positive impact on image analysis, both speed and accuracy, supporting diagnosis. In controlling the imager, itself, it has also been shown to have accelerated image collection rates, reducing procedure time and patient radiation exposure.

 

Obviously, AI is having a broad impact on the industry as a whole. And, in a cost constrained era, AI can be a difference-maker, adding more functionality and improving performance while potentially lowering operational costs. But only if it can be effectively integrated into the existing, very large, complex code bases of modern imaging systems without creating costly externalities.

 

Properly approached, it is possible to integrate new functionality into older codebases without introducing ripple effects that result in large program delays and even larger software costs. Using software and system architecture analysis techniques makes integration efficient and leads to improvement of the overall software system as well.

 

New Cutting-Edge Functionality, Twenty-Year Old Codebases

 

MRI, PET, CT and ultrasound imagers have exceedingly complex software control systems which are fully integrated with other parts of the imaging ecosystem, including image analysis and display, PACS, EMR systems as well as treatment planning systems. As discussed, rapidly evolving AI capabilities can significantly improve both the performance of the imaging equipment and the workflows for reading of the results. This represents one of the most difficult classes of technology to incorporate into existing medical equipment: fast-moving technology into a slower moving, very complex and highly regulated sector.

 

OEMs are being presented with difficult decisions regarding evaluating not only the current state of different AI engines, but also predicting the rate of improvement of the various AI technologies. Betting on the wrong engine could have very negative impacts on competitive position in the very near future.

 

This predicament requires a systems thinking approach and formalized architecture assessment. The challenge with AI (or any material functional addition) is to excise and replace core functionality in a manner that is compatible with the system’s design while minimizing implementation efforts.

 

The first step is to define the new AI functionality in a lexicon that aligns with the core system design. The new functionality must fit into the existing system within the context of the current architectural design. The AI capabilities should be integrated in an incremental manner; at first augmenting, then ultimately replacing current control and analytical functionality. By aligning the new functionality to the design concepts of the existing system it is usually possible to approach the redesign in a minimalist fashion.

 

The next step requires formally assessing the existing software system from an architectural perspective. Using techniques such as Architecture Tradeoff Analysis Methodology (ATAM), a complex software system can be decomposed to show key decisions, major components, interaction mechanisms and functional dependencies. This information can then be organized in a manner that allows a determination of the impact of system changes, highlighting approaches and tradeoffs that may be either supportive or inhibitive to inclusion of new functionality. This assessment will illuminate system risks and provide for the development of a roadmap to enhancing the system design allowing for integration of an AI module (or various AI modules) into the codebase, while developing a more structurally sound system design.

 

There are two direct side effects of this type of rigorous architectural analysis. The first is a more thorough understanding of hidden connections within the software system that create both constraints as well as accelerated design fatigue. This specific knowledge is critical in understanding system vulnerabilities to significant risks such as cyber-attacks or loss of critical patient-related data. Both of these are real-world challenges within the installed base of imaging systems.

 

The second benefit is reduced costs related to long-term support of the software system. Hidden connections and misunderstood architectural designs directly impact the ease and efficiency of making changes and provide more understanding of sources of instability within the current implementation. In an environment of more limited R&D spending, reducing the downstream costs of legacy support directly frees up monies to be invested in forward development.

 

Development and integration of new AI algorithms can be seen as a double opportunity for medical imaging manufacturers. The obvious potential of system improvement has gained widespread support. The new functionality will fundamentally change the industry. However, the implications of integrating such a different type of technology within the constraints of large, complex legacy software systems provides a second opportunity to systematically evaluate and improve the inner workings of the software architecture and design. It would be a missed opportunity to address one without the other.