AWS Partner Network (APN) Blog
Streamline Medical Imaging Workflows with Flywheel on Amazon S3
By Erin Chu, Director of Technical Sales – Flywheel
By Carl Luo, Partner Solution Architect – AWS
By Hailey D’Silva, Partner Solution Architect – AWS
By Saurabh Mishra, Sr. Partner Solutions Architect – AWS
 |
| Flywheel |
 |
Applying artificial intelligence and machine learning (AI/ML) in medical imaging has proven to be both powerful and successful. The FDA has approved nearly 700 AI/ML-Enabled Medical Devices. Of these devices, 77% are designed for radiological applications. Medical imaging is a common part of diagnostic plans.
In 2023, approximately 500 million medical imaging exams were performed on adults in the United States, with almost 2 medical imaging exams per adult. Medical data retention is mandated for at least 5 years in most states. Finally, medical imaging is time-and labor-intensive. Radiology technicians and radiologists undergo advanced training to capture and interpret medical image exams. While a simple X-ray may take under a minute, advanced imaging studies are more time-consuming. For example, Magnetic Resonance Imaging can take upwards of an hour.
Healthcare organizations face several critical challenges when leveraging their medical imaging data for machine learning applications. First, siloed data storage creates complexity – medical imaging data typically resides across multiple disconnected repositories within a healthcare system. Second, stringent compliance requirements mean organizations must carefully pre-process their imaging data, implementing robust deidentification, standardization, and harmonization workflows before feeding it into ML models. The sheer scale of medical imaging data compounds these challenges. With hospitals generating an average of 50 petabytes of data annually, of which approximately 80% comprises medical images, organizations need solutions that can handle massive datasets efficiently and securely.
Flywheel Solution Overview
Flywheel runs on AWS to deliver software as a service (SaaS). As shown in Figure 1, Flywheel builds on AWS services, and enables end-to-end processing of medical imaging data at scale. It seamlessly integrates with Amazon Simple Storage Service (Amazon S3) for secure, scalable storage and intelligent cost optimization.
Through this integration, healthcare organizations can manage, analyze, and share medical imaging data while reducing costs and time spent on data preparation. AWS customers have used Flywheel to achieve up to 94% cost reduction in medical imaging workflows, and reduce months-long manual processes to days.
Figure 1 – Flywheel Architecture Diagram
Amazon S3 scales effortlessly, handle growing data with performance. As with all health data, security is essential, Amazon S3 is HIPAA eligible and provides enhanced encryption features, access controls and audit logging to help maintain data integrity and compliance with regulations. Through Flywheel’s Interfaces tab, you can mount multiple Amazon S3 buckets onto Flywheel for data import, export, or both with a click of a button.
To enable secure access to your medical imaging data, you can connect your Amazon S3 bucket to Flywheel by configuring AWS credentials. First, specify your S3 bucket (for example, s3://my-flywheel-bucket) in the Flywheel interface. Then, provide the AWS Access Key and Secret Access Key associated with an AWS Identify and Access Management (IAM) role that has appropriate permissions to access the bucket. This configuration, illustrated in Figure 2, establishes a secure connection between Flywheel and your S3 storage.
Figure 2 – Add S3 Storage Provider
Once you have mounted your S3 bucket, you can import data into a Flywheel Project through the Web Import tool as shown in Figure 3. For this example, we’ve loaded one subject from the NSCLC-Radiomics-Interobserver1 project. This project is included in the Imaging Data Commons and is freely accessible from the Registry of Open Data on AWS.
Figure 3 – Import data into S3 Storage
Within a Flywheel project, Flywheel categorizes your data by Subject, Session, Acquisition, and File, as you can see in Figure 4. With Flywheel, you don’t need to look for different data types in different places, Flywheel helps you put it all in one place. For example, a patient may receive an X-Ray one visit, then an MRI the next, and complete survey saved as a PDF or a CSV. These data types will be filed under the same Subject in Flywheel.
Figure 4 – Console to view data
You can then visualize your data in Flywheel’s zero footprint radiology viewer as shown in Figure 5. The reviewer allows users to view, annotate, and segment medical images.
Figure 5 – View Health Image
Flywheel optimizes customer storage costs by transitioning data between access tiers based on usage patterns using Amazon S3 Intelligent Tiering. When using Reference-in-Place feature Flywheel doesn’t copy files, which reduces storage costs. Instead, Flywheel only reads the source files to extract metadata and stores a reference, thus treating the source location as primary storage.
Prepare data ready for analysis with Flywheel on AWS
Once your data has been imported into Flywheel, you can process and analyze with Flywheel Gears, leveraging configurable Amazon Elastic Compute Cloud (Amazon EC2) instances to optimize customer time and cost savings. If your Project has a Flywheel Gear Rule enabled, you can also automatically kick off data pre-processing. Our Example Project has a simple Gear Rule set up for two main purposes. First, it extracts Digital Imaging and Communications in Medicine (DICOM) header metadata. Second, it functionally classifies incoming DICOM files. The results of these processes are reflected in the Session dashboard that you see in Figure 3.
In addition, you can train machine learning models directly within the Flywheel platform with Flywheel Workspaces. Cohort building is simple using Flywheel’s Search function, where you can search on medical image attributes or on accompanying structured data.
Sharing your data with Flywheel is simple. You can share your data with collaborators within the Flywheel platform through Flywheel’s granular roles and permissions, or through the Flywheel SmartCopy feature. Alternatively, you could simply export a Flywheel project to Amazon S3 bucket for pickup.
By integrating Flywheel into your AWS solutions, you spend less money and time executing the undifferentiated heavy lifting of data aggregation, processing, and curation, and more time generating insights.
Empowering Healthcare with Flywheel and Amazon S3
The integration of Flywheel with Amazon S3 offers a powerful solution for streamlining medical imaging workflows. This combination enables healthcare organizations to efficiently manage, analyze, and share vast amounts of imaging data while significantly reducing costs and preparation time. As the volume of medical imaging data continues to grow, Flywheel on AWS provides a robust, scalable foundation for advancing AI-driven healthcare innovation.
To learn more about how Flywheel can help you make the most out of your medical imaging and associated health data, visit flywheel.io.
Flywheel – AWS Partner Spotlight
Flywheel is an AWS Advanced Technology Partner. Flywheel Enterprise is the revolutionary research data management platform powering healthcare innovation by accelerating collaboration, enabling machine learning, and streamlining the massive task of data aggregation, curation and management.