Revolutionizing healthcare with AI-driven digital pathology

In the world of modern medicine, pathology delivers the definitive diagnoses that guide critical treatment decisions. For over a century, pathologists have peered through microscopes at glass slides—a method that, while effective, has remained largely unchanged despite technological advances elsewhere in healthcare.

Today, a powerful transformation is underway. Digital pathology converts traditional glass slides into high-resolution whole slide images (WSIs) that can be analyzed anywhere, anytime. It eliminates the physical constraints of microscope-based review, while enabling faster diagnosis and improved patient outcomes.

However, the true revolution comes when artificial intelligence meets digital pathology. AI algorithms can automatically identify tissue structures, detect cellular abnormalities, and quantify biomarkers with remarkable precision, often exceeding human consistency. Machine learning (ML) models can classify tissue types, help grade tumors, and highlight areas of concern, while natural language processing generates structured reports that seamlessly integrate with clinical protocols. This intelligent diagnostic pipeline reduces pathologist workload while ensuring human clinical judgment remains central.

This transformation couldn’t come at a more critical time. Healthcare systems worldwide face mounting challenges:

• Pathologist shortages
• Rising cancer cases
• Growing demand for standardized diagnostic workflows

Amazon Web Services (AWS) is uniquely positioned to help clinicians address these challenges with our purpose-built healthcare and AI services, alongside a robust portfolio of pathology partner solutions. Together, we’re helping healthcare organizations transform traditional pathology into intelligent diagnostic pipelines, all while maintaining the essential supervision and expertise of pathologists.

Now, let’s explore a solution that can:

• Ingest and store digital pathology images using AWS HealthImaging, a DICOM-compliant medical imaging storage solution. It provides sub-second access to pathology images with healthcare-specific security and compliance features.
• Train and deploy ML models to perform inference on pathology images using Amazon SageMaker AI, which provides fully managed infrastructure, tools, and ML workflows.
• Build and deploy AI agents to augment pathology report generation and validation processes using Amazon Bedrock and Amazon Bedrock AgentCore.

HEALTHCARE AI MODEL DISCLAIMER: These healthcare AI models are for research and development purposes only. Model performance has not been validated for medical applications and should not be deployed in clinical settings or used for diagnosing or treating health conditions. Users assume full responsibility for verifying all outputs, ensuring compliance with healthcare regulations, obtaining required regulatory approvals before incorporation into medical products/services, and following applicable clinical decision-making protocols.

Digital pathology architecture on AWS

The following architecture showcases the entire process from slide digitization through final report delivery, incorporating AI-powered image analysis, automated report generation, and seamless integration with existing healthcare data systems. Pathologists remain central to the workflow, with AI serving as an intelligent assistant that enhances clinical expertise.

Figure 1 – AWS-powered digital pathology workflow

How it works

Let’s dive deeper into the steps highlighted in the architecture to understand how AWS services can be leveraged to revolutionize the digital pathology workflow with AI.

1 – Sample collection and processing

Physicians create a pathology order in the electronic health record (EHR) system and collect the biopsy sample from the patient. The sample is then transported to the pathology lab to undergo standardized histopathology preparation including fixation, embedding, sectioning, and staining. This controlled process confirms proper sample identification and maintains chain of custody for regulatory compliance. It also creates consistent, high-quality specimens optimized for digital imaging and AI-powered analysis.

2 – Digital slide creation and storage

Following the processing steps, stained slides are created which undergo digitization using advanced whole slide imaging scanners that capture gigapixel-resolution images with multiple magnification levels. These high-resolution digital slides are stored in AWS HealthImaging, enabling sub-second time-to-first-image access that enables high-throughput pathology workflows and near real-time diagnostic review.

The service provides enterprise-scale interoperability through purpose-built medical imaging storage and provides standardized APIs designed for healthcare interoperability. This includes DICOMweb APIs that enable seamless integration with existing digital pathology viewers, picture archiving and communication system (PACS), and third-party applications.

Pathology data stored in AWS HealthImaging serves as an AI data foundation, providing accessibility for AI development and machine learning workflows. Additionally, AWS HealthImaging streamlines digital pathology migrations with support for legacy whole-slide imaging archives converted to the DICOM standard. It reduces the IT burden, while maintaining data continuity, and enables laboratory interoperability with vendor-neutral storage in the cloud.

3 – Model training and inference

ML models deployed through Amazon SageMaker AI endpoints perform inference on whole slide images to provide automated insights that augment pathologist expertise. These trained pathology foundation models can perform a variety of sophisticated tasks. The tasks can include tissue segmentation, cell counting, biomarker quantification, and anomaly detection, delivering results in near real-time to support clinical decision-making. SageMaker AI also offers access to hundreds of pretrained models, including publicly available foundation models, that can be fine-tuned and deployed in a few quick steps.

Models can be invoked through an event driven architecture from the digital pathology solution hosted on AWS. It would automatically happen when images are ingested from the scanners into AWS HealthImaging.

The model training environment leverages Amazon SageMaker Training for distributed training of large pathology foundation models using GPU clusters. This enables the development of sophisticated pathology models capable of understanding complex tissue patterns and cellular structures. Amazon SageMaker notebook instances provide interactive development environments where researchers and data scientists can experiment with model architecture and validate performance against clinical datasets. SageMaker AI supports governance requirements with streamlined access control and transparency regarding AI projects.

AWS HealthImaging enhances training workflows by storing pixel data and metadata separately with independent access controls. Researchers can compose training datasets using only pixel data while maintaining appropriate security boundaries for sensitive clinical information.

Amazon Simple Storage Service (Amazon S3) serves as the backbone for storing massive pathology datasets that can be imported from AWS HealthImaging for training purposes. Amazon S3 has comprehensive versioning and metadata management capabilities that support reproducible research and model development workflows.

4 – Pathologist review

Following model inference, the digital pathology system updates case status to notify pathologists that results are available for review. Pathologists access results through digital viewers retrieving image data stored in AWS HealthImaging. Pathologists can compare AI-generated findings against their clinical expertise, while conducting a thorough examination of flagged regions.

This collaborative workflow maintains pathologist control over final diagnostic decisions. They can leverage AI assistance to accelerate routine tasks and highlight areas requiring detailed investigation, facilitating both diagnostic accuracy and workflow efficiency. The pathologist findings, alongside the findings from the AI models that the pathologist would like to use, are stored in the digital pathology solution. They are then used by the Report Generation and Report Validation agents in the next step.

5 – Report generation and validation

After the pathologists have completed their review, they can trigger the report generation and validation workflow from the digital pathology solution. It uses specialized AI agents deployed through AgentCore Runtime, provided by Amazon Bedrock AgentCore, which is a secure, serverless runtime purpose-built for deploying and scaling dynamic AI agents. The AgentCore Runtime manages the agent lifecycle, inter-agent communication, and workflow state management. It enables specialized AI agents to collaborate on complex diagnostic processes while maintaining coordination and consistency across all operations.

The agents used in this workflow are:

• Pathology Reporting Orchestrator: Receives requests from pathologists through the digital pathology system and coordinates the sequential execution of report generation and validation steps. It manages the workflow state and verifies proper handoffs between the Report Generation and Report Validation agents.
• Report Generation agent:  The Pathology Reporting Orchestrator invokes the Report Generation agent, which analyzes findings from Step 4 to generate pathology reports using Amazon Nova Lite large language model (LLM) through Amazon Bedrock. Amazon Bedrock provides model customization capabilities for specific models such as Amazon Nova Lite which customers can choose to perform for more accurate report generation results. The agent incorporates quantitative measurements and diagnostic findings into standardized templates while querying the Amazon Bedrock Knowledge Bases through Retrieval-Augmented Generation (RAG) capabilities for current medical guidelines.
• Report Validation agent: After the report is generated by the Report Generation agent, the Pathology Reporting Orchestrator invokes the Report Validation agent. It performs systematic quality assurance by reviewing the generated reports against clinical standards and protocols using Amazon Nova. The agent validates the report’s accuracy, completeness, and CAP guideline adherence by cross-referencing content with pathology protocols stored in the Amazon Bedrock Knowledge Bases through RAG capabilities. It also flags inconsistencies or missing information before reports enter the clinical workflow.

6 – Preliminary pathology report generation

Upon completion of the agentic workflow, a preliminary pathology report alongside validation results becomes available for pathologist review. Pathologists can examine the AI-generated report and validation results—making final changes as needed before submitting the report, which will update the case status in the Laboratory Information System (LIS) system. This status update is communicated to the EHR system through interfaces setup between these two systems making the final report available to the physician for clinical decision purposes.

7 – Patient diagnosis and treatment

As the final step in the workflow, the physician will review the final report to understand the diagnosis of the patient and create a treatment plan, which is communicated to the patient.

Conclusion: Accelerating the Future of Digital Pathology with AWS

Whether you’re a cutting-edge research institution, a healthcare organization serving patients, or an AI model provider developing next-generation algorithms, AWS provides the foundation to help transform your digital pathology vision into reality.

The AWS solution outlined delivers what modern healthcare demands: comprehensive security and compliance frameworks that protect sensitive patient data. It also provides high-performance computing that scales with your imaging needs, and enterprise-grade reliability that confirms diagnostic workflows remain uninterrupted. This robust infrastructure can facilitate your digital pathology implementations needs while exceeding today’s stringent regulatory requirements.

Beyond our technology, AWS brings together an extensive network of medical imaging partners and AI algorithm vendors specializing in digital pathology. These strategic partnerships empower you to select from proven, field-tested solutions, while accelerating implementation through established integrations and specialized expertise.

As pathology continues its evolution from microscopes to machine learning, AWS stands ready to support healthcare’s most critical diagnostic discipline. By combining the irreplaceable clinical judgment of pathologists with the power of digital transformation, and artificial intelligence, together we’re building a future. Diagnoses can be made faster, with more accuracy, and ultimately lead to better patient outcomes—the true measure of success in healthcare innovation.

Ready to transform your pathology workflows? Contact an AWS Representative to find out how we can help accelerate your business.

Further Reading

• Accelerate digital pathology slide annotation workflows on AWS using H-optimus-0
• AWS Helps Genomics England’s Multimodal Programme Accelerate Research with Whole Slide Images
• Enhancing Multidisciplinary Collaboration in Digital Pathology with Cloud-Based PACS
• Unlocking the Power of the Cloud for Digital Pathology: How Proscia and AWS Deliver a Purpose-Built SaaS Platform