AWS for Industries
Accelerate breast cancer treatment planning with agentic AI
In the complex world of oncology, precision and speed are paramount. From mammogram to personalized treatment plan, artificial intelligence is revolutionizing every step of the breast cancer diagnostic journey. Let’s explore how emerging agentic AI technologies are transforming this critical decision-making landscape, potentially enhancing treatment acceleration while addressing real-world implementation challenges.
The clinical challenge: From diagnosis to treatment plan
Following a breast cancer diagnosis, the clinical path transitions from biopsy to comprehensive treatment planning. Oncologists must face the intricate process of synthesizing complex medical information, research insights, and clinical guidelines to chart the most effective therapeutic path.
One of the most significant challenges oncologists face is aggregating and interpreting multimodal data from various sources:
- Imaging studies
- Pathology reports
- Genome sequencing
- Clinical history
- Real-time patient monitoring
The sheer volume and diversity of this data can be overwhelming. Interoperability issues between different health information systems may further complicate creating a unified view of the patient’s condition. Staying current with rapidly evolving research also introduces additional complexity.
We propose an agentic solution engineered for the automated and comprehensive extraction and summarization of heterogeneous data sources. Sources include multimodal medical encounter data, longitudinal patient histories, and the latest clinical guidelines. The synthesized output from this agentic framework supports enhanced clinical decision-making by providing a consolidated basis for oncologists to formulate evidence-based, personalized therapeutic treatment plans. This approach fundamentally optimizes the clinical workflow, by offloading resource-intensive data consolidation tasks, and aids in treatment planning.
However, for our purposes here we are going to be focusing on the agentic AI portion of the full solution.
The revolutionary promise of agentic AI
Healthcare is entering a transformative phase with the emergence of agentic AI, which goes far beyond conventional AI technologies.
Unlike traditional systems, agentic AI demonstrates:
- Autonomous decision-making capabilities
- Adaptive learning in complex healthcare environments
- Real-time monitoring of patient progress
- Dynamic coordination of care teams
- Accelerated treatment plan development
With their unique blend of innovative problem-solving and real-time adaptability, agentic AI solutions can handle both routine administrative tasks and intricate data management. They can free up healthcare providers from different burdens so they focus more on direct patient care and crucial medical interventions.
Solution architecture: A multi-agent approach
Here, we implement a multi-agentic orchestration framework for accelerating clinical decision support. This design pattern represents an async workflow management approach in multi-agent systems. A central orchestrator dynamically coordinates complex tasks by intelligently planning, decomposing, and delegating subtasks to specialized worker agents.
Unlike traditional rigid workflows, this pattern offers flexibility, allowing subtasks to be determined contextually based on specific input rather than predefined sequences.
The architecture uses a distributed problem-solving approach through a multi-agentic orchestration framework with pathology, imaging, and clinical data assembled into AWS HealthImaging and AWS HealthLake services.
Our innovative agentic AI solution is built on generative AI technologies using managed services such as Amazon Bedrock, Strands Agents, and Amazon Bedrock AgentCore. This solution is engineered for automated and comprehensive extraction and summarization of heterogeneous data sources, supporting enhanced clinical decision-making.
Following are the key agents used within the solution:
- Treatment Planning Orchestrator agent (central coordinator): Coordinates the entire workflow, processes user prompts and dispatches requests to specialized sub-agents. It can synthesize information to generate treatment options.
- Patient Multimodal Data Synthesizer agent: Gathers comprehensive current encounter data and accesses clinical and imaging data from AWS HealthLake and HealthImaging APIs.
- Relevant Clinical History Extractor agent: Extracts relevant patient history from various tools and identifies conditions (such as hypertension) that may impact treatment decisions.
- Evidence-based Guidelines agent: Invokes tools for American Society of Clinical Oncology (ASCO) and National Comprehensive Cancer Network (NCCN) guidelines, and pulls latest breast cancer information relevant to the patient’s case. It also extracts relevant clinical trials from clinicaltrials.gov, and researches new breast cancer treatments from PubMed.
Throughout this process, the oncologist maintains control through a human-in-the-loop approach, reviewing and refining the AI-generated recommendations before final approval.
The solution also leverages several AWS services for various functions:
- AI agents implementation: Built using Strands Agents and powered by Claude 3.7 Sonnet model from Anthropic on Amazon Bedrock, the evidence-based guidelines agent uses the native retrieve tool of Strand Agents to access NCCN and ASCO guidelines stored in Amazon Bedrock Knowledge Bases.
- Amazon Bedrock AgentCore Runtime: Purpose-built for agentic workloads, it provides extended runtime, fast cold starts, and true session isolation. It supports built-in identity and multi-modal payloads. Deployment involves configuring an entry point, which generates a Docker file, creating a container image and pushing to Amazon Elastic Container Registry (Amazon ECR).
- Amazon Bedrock AgentCore Gateway: Enables secure discovery and access to tools, supporting multiple input types including OpenAPI, Smithy, and AWS Lambda functions. Implements comprehensive authentication protocols and uses Lambda functions as Model Context Protocol (MCP) tools.
- Security and compliance: Authentication is done using Amazon Cognito with OAuth2 machine-to-machine authentication, which is designed to meet HIPAA compliance requirements. AI agents operate within an MCP client with unique gateway access tokens, facilitating granular access control and traceability.
- Data access: Through Lambda functions API calls, secure access is granted to patient clinical data from AWS HealthLake, and medical imaging data from AWS HealthImaging. Research data from medical research databases and clinical trials data are accessed through separate Lambda functions using public APIs.
The full solution
Complete code and instructions for implementation are available within our GitHub repository. While we have focused on breast cancer for our demonstration, this agentic solution can be applied to any type of treatment planning.
Note, before beginning the implementation of this solution, you’ll need:
- AgentCore starter toolkit for fast setup
- AWS SDK for Python (Boto3)
- Strands Agents SDK
- Amazon Bedrock
- AWS Command Line Interface (AWS CLI)
Figure 2 is what the solution looks like when implemented and running. It demonstrates the oncologist’s interaction with the tool and how the agentic AI solution responses can aid in treatment plan building.
The future of oncology care
The integration of agentic AI within clinical workflows marks a transformative milestone in oncology patient care. By intelligently reducing administrative burdens, these AI systems enable oncologists to focus on patient-centered care and nuanced medical judgment.
The human-in-the-loop design verifies that AI serves as a powerful augmentation tool rather than a replacement for clinical expertise. It provides evidence-based recommendations while preserving the irreplaceable human elements of medical decision-making.
As healthcare continues to generate unprecedented volumes of data and complexity, agentic AI stands poised to become an essential technology for delivering personalized, precise, and scalable medical interventions. The future of medical care is collaborative, with AI and human expertise working in harmonious synergy to unlock unprecedented potential in diagnosis, treatment, and patient outcomes.
Ready to transform your clinical decision support? Explore the Amazon Bedrock AgentCore and the Strands Agents SDK documentation and discover how AI can revolutionize your organization’s approach to providing decision support for oncologists. Contact an AWS Representative to find out how we can help accelerate your business.