AWS for Industries
Revolutionize personalized radiology learning using AI and AWS
Radiology training stands at a crossroads. While imaging volumes surge and diagnostic complexity increases exponentially, the traditional apprenticeship model struggles to deliver comprehensive education without compromising patient care delivery. How can radiology educators deliver thorough and personalized instruction to users without impacting their clinical responsibilities? Also, can such a personalized radiology education be scaled to reach users who are anywhere in the world?
The League of Radiologists (LOR), developed at Massachusetts General Hospital (MGH) and Harvard Medical School (HMS), offers an answer. Built on Amazon Web Services (AWS), this platform integrates AI, adaptive learning, and gamification to transform radiology training into an interactive, scalable experience. It meets users wherever they are on their educational journey.
The challenge: scarcity of trustworthy datasets and scalable training
Radiology training traditionally relies on side-by-side review of clinical cases complemented by independent study of static case libraries. This method is limited by faculty time for teaching, the relatively small number of clinical cases reviewed, and the inability to adapt to each learner’s specific needs.
A central barrier to AI integration is the scarcity of high-quality radiology datasets curated specifically for pedagogy. Most large imaging archives contain label noise or diagnostically complex, multi-finding cases that are unsuitable for foundational learning. Healthcare institutions need solutions that can systematically transform vast collections of clinical images into educational content tailored to different training levels, and clinically relevant to real-world radiology practice.
The solution: AI-powered adaptive framework
The League of Radiologists platform (Figure 1) reimagines radiology education through intelligent automation and cloud scalability.
The LOR framework integrates three synergistic components to overcome the data quality and scalability challenges:
1. Data curation: A hybrid AI-human expert pipeline rigorously filters noisy public archives to establish a foundation of trustworthy, single-finding cases suitable for foundational learning.
2. Content generation: A large language model (LLM) pipeline automatically generates a rich library of clinically grounded questions designed to build five core radiology reasoning skills.
a. Finding detection: Determine whether the radiograph shows no abnormality or contains a reportable finding.
b. Diagnosis: Identify the most clinically significant finding in the image.
c. Finding attributes: Assess a specific attribute of the finding, such as its anatomical location or descriptive characteristics.
d. Finding identification: Select the correct finding shown within the highlighted bounding box.
e. Bounding box drawing: Draw a bounding box around a specified finding in the image.
3. Adaptive delivery: The content is deployed on an interactive, gamified platform that uses an adaptive algorithm to deliver a personalized and engaging learning experience.
Figure 1: League of Radiologists platform login
The system tailors questions to each user’s current skill level and continuously adjusts the difficulty based on performance (Figure 2). Gamification elements, including progress milestones and achievement tracking, maintain motivation throughout the learning journey (Figure 3).
Figure 2: Sample chest radiograph question
Figure 3: Achievement tracking
Architecture: Building intelligence on AWS
LOR leverages AWS managed services to deliver scalable, intelligent, and reliable education:
- Amazon Bedrock powers the platform’s ability to generate thousands of unique, clinically accurate questions. It analyzes real radiology reports and creates realistic diagnostic scenarios that mirror actual practice. The service provides access to multiple foundation models, while maintaining HIPAA eligibility, essential for processing medical data. AI agents can furthermore be leveraged through Amazon Bedrock Agents and Amazon Bedrock AgentCore. In LOR, multiple specialized agents work together to extract clinical findings, generate questions, and create realistic answer options that reflect real-world diagnostic challenges.
- Amazon Comprehend Medical extracts medical terminology and relationships from unstructured radiology text, enabling generated questions to align with authentic clinical language and scenarios.
- Amazon Neptune builds a knowledge graph connecting radiology concepts, linking imaging findings, diagnoses, and anatomy to improve contextual understanding and question relevance.
- Amazon SageMaker enables adaptive learning by training and deploying models that adjust question difficulty in real time based on individual user performance patterns.
- Amazon CloudWatch monitors the platform’s performance and quality metrics.
- Amazon Augmented AI (A2I) enables human expert review for quality assurance and oversight.
This cloud-based design enables global reach, automatic scaling, and real-time analytics that help educators understand the user’s progress, while quickly refining content.
Results: Validated impact on pedagogical reliability
The LOR platform’s initial chest radiography module demonstrated impressive scale: starting with 493,785 images, the platform produced 2,305 validated multiple-choice questions from 881 high-confidence cases. The platform has also been deployed successfully as a publicly accessible learning tool.
Key outcomes included:
- Clinical and pedagogical accuracy: Expert validation confirmed that the content met the high standards required for medical education. The hybrid curation pipeline (combining expert-labeled cases with an AI-verified cohort filtered at a 100% Positive Predictive Value) confirms instructional reliability.
- Improved engagement: Gamified learning and personalized progression increased user participation and retention compared to traditional methods.
- Structured reasoning: Questions were mapped to five core reasoning categories (Finding detection, Diagnosis, Finding attributes, Finding identification, and Bounding box drawing) to guide the user through the entire diagnostic process.
- Global accessibility: The platform reached users across multiple countries, eliminating barriers, such as specialized software or hardware requirements.
This approach demonstrates how AWS AI and analytics services enable personalized medical education to scale, while maintaining rigorous clinical standards. By automating content curation and question creation, the LOR framework provides educators with a way to shift their focus from manual content generation to mentorship.
Lessons for healthcare innovation
This experience at MGH and HMS offers insights for other healthcare organizations exploring AI-powered education:
- Educational reliability requires hybrid validation: Domain-specific data curation combined with a hybrid validation strategy (automated checks, user feedback, and expert review) is essential to maintain clinical soundness, while providing scalability.
- Data curation is paramount: The persistent challenge of label noise in clinical archives must be addressed. Our process of restricting inclusion to single-finding cases created a reliable foundation for core learning.
- Gamification must reward competency: The most effective approach rewarded competency growth through adaptive difficulty adjustment, rather than activity metrics.
- Design supports global audiences: International users benefited from clear medical language, defined as standardized global terminology rather than local jargon, and an infrastructure that performs consistently across regions.
What’s next
Building on the success of chest imaging, the team plans to expand LOR to additional specialties including neuroimaging, musculoskeletal imaging, and abdominal imaging. Future versions will incorporate multimodal AI models that combine text and imaging data for deeper clinical understanding.
The vision is a globally accessible, AI-powered radiology education ecosystem that continuously learns, adapts, and improves alongside the healthcare professionals it serves.
Why AWS
AWS provides the capabilities that transformed LOR from concept to reality:
- Scalability: The platform handles growing user demand automatically, without manual infrastructure management.
- Security and compliance: HIPAA-eligible services including Amazon Bedrock and Amazon SageMaker protect sensitive health information.
- Innovation velocity: Managed services eliminate infrastructure concerns, providing the team with a way to focus on improving educational outcomes.
With AWS, this initiative at MGH and HMS transformed a complex AI vision into a practical, scalable solution that improves radiology education and, ultimately, patient care.
Conclusion
The League of Radiologists platform, developed at Massachusetts General Hospital and Harvard Medical School, created an AI-powered, scalable educational gamification for radiology training. It adapts to the user’s level of knowledge, increasing the difficulty of real world-based questions, to help push the user as they continue to learn.
The platform offers rewards and encouragement along the way to promote the continued user’s growth. This personalized instruction is done without impacting the user’s clinical responsibilities and has been scaled to reach users who are anywhere in the world.
The LOR platform has increased clinical and pedagogical accuracy through its structured reasoning to a global audience of users.
To explore the League of Radiologists platform visit radontology.org or contact an AWS Representative to find out how we can help accelerate your business.
Where to learn more
Healthcare organizations and educators interested in developing similar scalable learning solutions can explore AWS healthcare AI services:
- AWS Health Data Portfolio site
- AWS for Healthcare & Life Sciences
- AWS Healthcare Solutions
- You can also read more blogs about AWS healthcare stories

