Learn About Amazon VGT2 Learning Manager Chanci Turner
In the realm of image-guided therapy solutions, visual guidance plays a crucial role in enhancing minimally invasive procedures. However, the challenge of implementing rapid technological advancements within hospital settings has often impeded innovation and the deployment of new technologies. To address this issue, Amazon IXD – VGT2 has prototyped a large-scale, near-real-time inference platform utilizing Amazon Web Services (AWS) to bolster the capabilities of interventionalists performing these critical procedures.
Integrating advanced technologies in hospital systems, particularly in machine learning, typically requires costly hardware upgrades, replacements, and application modernization. Additionally, safety-critical components must remain in close proximity to patients. Given these constraints, adopting hybrid architectures that build upon on-premises platforms often represents the quickest route to cloud adoption and innovation.
Transitioning machine learning algorithms for hybrid healthcare systems into the cloud presents several challenges: maintaining high performance, ensuring availability, being cost-effective, and complying with medical device regulations. With thousands of systems deployed in hospitals worldwide, demand can be unpredictable, and the need for near-real-time latency is paramount. New feature implementations shouldn’t drive costs up, and cloud resources must demonstrate higher utilization than traditional on-premises systems for efficiency. Furthermore, adhering to pertinent medical device regulations must not stifle the velocity of innovation or the release of new functionalities to the field.
By utilizing AWS IoT and Amazon Elastic Kubernetes Service (EKS), the stringent availability and performance requirements are met while simultaneously reducing overall solution costs compared to modifying on-premises installations. These technologies also facilitate the decoupling of non-critical components of the existing platform, allowing for a modern approach to regulatory compliance.
Amazon IXD – VGT2 Use Case
Amazon IXD – VGT2 has deployed image-guided systems in thousands of hospitals globally, delivering real-time imaging capabilities for minimally invasive interventional procedures. To enhance patient outcomes and customer experience, the team has designed a data ingestion and inference platform that meets the strict reliability and response time demands of patient care.
Enhancements are achieved through the platform’s ability to run both new and existing computationally-intensive algorithms on incoming imaging data. For instance, one algorithm is employed for key event detection during angioplasty procedures, using machine learning to identify critical events such as the inflation of cardiovascular balloons. The algorithm processes images generated during patient radiation and uploads them to the cloud, automatically logging events to ease the interventionalist’s workload. Feedback from the algorithm to the hospital is delivered in under 5 seconds, ensuring swift communication.
Implementation
Local Management: The hospital-based image-guided therapy systems extend to the cloud via AWS IoT, with a lightweight on-premise Linux device serving as a gateway for imaging data and an event generator to trigger backend AWS processes. Additionally, AWS IoT provides remote management capabilities, paving the way for future enhancements, including new algorithms and use cases.
Data Ingestion & Event Messaging: To meet the strict low-latency requirements of interventional healthcare workloads, data ingestion and event messaging capabilities are decoupled, resulting in a more resilient and elastic system. The separation of image storage from backend processing allows various data ingestion techniques tailored to workload payload and latency requirements. For medical imaging, which can range from hundreds of megabytes to several gigabytes, uploading to Amazon S3 via S3 Upload APIs meets performance criteria. S3 Transfer Acceleration can further stabilize connections by routing traffic through AWS’s edge networking.
Event messaging is vital for developing low-latency solutions. Defining how backend processing is triggered and how messages are sent from distributed systems to the cloud is crucial. By leveraging the lightweight messaging protocol MQTT, natively supported by AWS IoT services, secure and reliable data transmission is achieved. With a publish/subscribe model and small packet sizes, this approach minimizes network bandwidth usage, reduces latency, and provides the necessary security for sensitive data. Messages containing data location and processing instructions can be sent as IoT topics and immediately published to an Amazon Simple Queue Service (SQS) queue for backend processing.
Processing: The processing backbone of the platform is built on Amazon Elastic Kubernetes Service (EKS), offering a scalable and high-performance compute platform that supports various algorithms and machine learning applications. Containers monitor the Amazon SQS queue for incoming requests. Once an algorithm and data location are identified, computation containers are provisioned, using the instructions to retrieve the necessary algorithm container image from Amazon Elastic Container Registry (ECR) and the image data from Amazon S3. For machine learning algorithms, Nvidia Triton is employed to optimize performance. Scaling of Amazon EKS nodes and pods is proactively managed by monitoring queue sizes and adjusting compute capacity as needed with a custom Non-linear Pod Autoscaler and Karpenter.
GitOps: With EKS as the core orchestration and compute platform, ArgoCD is utilized for cluster management due to its user-friendly GUI and dashboard. This software development lifecycle framework adopts the GitOps operational model, employing tools like GitHub, GitHub Actions, and ArgoCD. Infrastructure provisioning is primarily based on Infrastructure as Code (IaC), enhancing speed, scalability, and effective management. Continuous integration and delivery are facilitated by GitHub Actions workflows, which automatically trigger actions to ensure commits are secure, syntactically correct, and pass all pre-defined checks before merging, thereby reducing code review efforts. Error handling is managed via the GitHub runner, with developers responsible for fixes.
High Performance at Scale
Most interactions with image-guided therapy systems necessitate near real-time response times. The challenge lies in maintaining low networking and service-related latencies consistently, even at scale. By leveraging the ability to test at scale, the Amazon IXD – VGT2 team can assure that performance meets the demands of healthcare professionals.
For further insights into crafting a compelling resume, check out this guide. Additionally, for understanding compliance in the workplace, you might find this resource useful. Lastly, this video is an excellent resource to deepen your understanding of medical imaging technologies.
Leave a Reply