Accelerating radiology imaging workflows with relevant clinical context on AWS

December 12, 2022 Priya Padate

Having ready access to clinical context helps radiologists eliminate assumptions and better apply their skills and expertise in rendering final reports. An automated extraction of relevant clinical context from electronic health records (EHRs) can provide clinical context.

A typical imaging workflow involves a patient following up for a recommended procedure prescribed by a provider. For an example, completing a CT scan procedure. The technologist performs the scan and verifies the images, which are then acquired into the picture archiving and communication system (PACS). This will be followed up on with a radiologist reporting and interpreting the study.

At the time of the reading, radiologists need more background on a patient’s medical history to make a narrower and more definitive diagnosis. Radiologists want to compare the current images against previously diagnosed and published cases. They also may need patient clinical content in context by searching through to EHR data such as surgical notes, pathology reports, and clinical notes.

Healthcare organizations utilize numerous applications―often with their own databases and different data models creating silos of data within the organization. An incomplete view of the patient record, with siloed data systems, requires the radiologist to search many systems in order to piece together the complete picture of the patient’s situation. These investigative activities are time consuming—impacting the overall capacity of the radiologist.

The value of the resulting radiology report is diminished by the lack of opportunity to review related clinical information.

Solution Architecture

Creating a diagnostic workflow automation during the interpretation process can support radiologists by presenting them relevant clinical context automatically based on current study attributes like:

  • Modality
  • Reason for exam
  • Procedure
  • Body part

This eliminates the need to manually search patient records, such as the EHR, as well as search comparison databases for more difficult or abnormal cases.

With Amazon Web Services (AWS), you can create a central location of your imaging metadata and clinical data. Then in a cloud-centric healthcare application, utilizing artificial intelligence (AI)/machine learning (ML) models, can fetch relevant context-based documents.

For a high-level design, start with the foundation of Amazon HealthLake. HealthLake takes data from diverse sources like EHRs, PACS, vendor neutral archives (VNAs) and creates a Fast Healthcare Interoperability Resource (FHIR) compliant data store to provide a complete view of each patient’s medical history. Using ML models, HealthLake automatically understands and extracts meaningful medical information from the raw data, such as medications, procedures, and diagnoses.

HealthLake can even map this information to medical ontologies for you at the same time, saving many hours of manual work. By leveraging these features, healthcare organizations can build applications that can improve:

  • Healthcare workflows
  • Clinical research
  • Healthcare informatics
  • Population health management
  • And more


Figure 1 – Clinical documents to Amazon HealthLake data store Workflow

Figure 1 – Clinical documents to Amazon HealthLake data store Workflow

The workflow starts with a patient encounter resulting in clinical data generated by the EHR systems for the visit.

  • The Clinical data could be clinical notes, surgery notes, labs, and more which are unstructured documents stored on Amazon Simple Storage Service (Amazon S3) in raw or transformed formats.
    • Amazon S3 is a HIPAA-eligible object storage service that is highly durable and scalable for storing large volumes of object level data. Amazon S3 has features targeting different storage types and price points for structured (for example, claims data or EHRs) and unstructured (radiology images, audio/video recordings, reports, and more) data at any scale. The data can be queried by downstream applications.
  • Amazon S3 triggers an event to run an AWS Lambda function when a new document is stored. The Lambda function generates a JSON file representing a FHIR DocumentReference resource.
    • AWS Lambda is a compute service that lets you run code without provisioning or managing servers. Lambda functions can be used for building data-processing triggers for services like Amazon S3 and to create and process FHIR resources. Since you pay only for the compute time you consume it’s very economical.
  • The generated DocumentReference FHIR resource is than submitted to Amazon HealthLake. To submit the DocumentReference in HealthLake, a standard FHIR REST API will be used to create the new resource.
  • Amazon HealthLake’s integrated use of Amazon Comprehend Medical extracts the medical terminology and augments the DocumentReference resources with attributes it discovers when a new DocumentResource is created. These new attributes are automatically appended to the DocumentReference in an extension attribute.
    • Comprehend Medical is a HIPAA-eligible natural language processing (NLP) service that uses machine learning (ML) to extract clinical data from unstructured medical text or medical concepts from written text found in clinical documents. Attributes such as conditions, diagnosis, International Classification of Diseases (ICD) and RxNorm codes, anatomical sites, and other information can then be surfaced back to the clinician—saving them time.

Figure 2 – Fetching relevant clinical and Imaging context

Figure 2 – Fetching relevant clinical and Imaging context

The Imaging workflow starts with the patient arriving for the scheduled procedure, the technologist performs the scan, verifies the images for quality and then the images are ready to be acquired on PACS.

  • The images are stored in an Amazon S3 bucket, created per department.
  • Amazon S3 triggers an event to run a Lambda function when a new document is stored. The Lambda profiles the study and the patient information, then persists the metadata on the Amazon Relational Database Service (Amazon RDS).
    • Amazon RDS is a collection of managed services that makes it straightforward to set up, operate, and scale databases. The biggest benefits are smooth scaling (in comparison to on-premise databases) as patient data grows and out-of-box encryption at rest and in-transit, which is crucial for healthcare data.
  • Another Lambda function is triggered to generate a JSON file representing a FHIR ImagingStudy resource. The generated ImagingStudy FHIR resource is than submitted to Amazon HealthLake using a standard FHIR REST API to create the new resource.
  • At this point the entire medical history for the patient (clinical and imaging) is available on HealthLake and can be associated by patient ID.

The Workflow:

When the radiologist pulls up the study for interpretation and reporting, the entire medical history for the patient (including imaging and clinical) is presented in a single patient jacket. We can further boost the workflow by presenting the relevant studies/documents for the current study. Based on relevancy detection we can prioritize the relevant imaging studies and clinical documents of the current study in context.

For imaging studies, relevancy detection can be done by matching historical studies modality, body part, reason for exam or procedure description (which are part of the ImagingStudy Resource attributes) with the current study in context.

For clinical documents, HealthLake automatically integrates with natural language processing (NLP) for the DocumentReference resource type. The integrated medical NLP output is provided as an extension to the existing DocumentReference resource.

The integration involves reading the text data within the resource, and then calling the following integrated medical NLP operations:

  • DetectEntitiesV2
  • InferICD10-CM
  • InferRxNorm

The response of each of the integrated medical NLP APIs is appended to the DocumentReference resource as an extension that is searchable.

In addition, for detecting Systematized Nomenclature of Medicine-Clinical Terms (SNOMED CT) code we can leverage Amazon Comprehend Medical. Amazon Comprehend Medical can identify clinical terms and provide ordered SNOMED CT concepts and code that might be referred to in the text. This is well suited for the following scenarios:

  • Assistance for professional medical coding in patient records
  • Clinical studies and trials
  • Population health management

SNOMED CT provides a comprehensive vocabulary of medical concepts, including medical conditions and anatomy, as well as medical tests, treatments, and procedures. It detects entities in the following categories:

  • Medical Condition—the signs, symptoms, and diagnosis of medical conditions
  • Anatomy—the parts of the body or body systems and the locations of those parts or systems
  • Test treatment procedure—the procedures that are used to determine a medical condition

Extract and Match concepts with Amazon Comprehend Medical SNOMED CT
Let’s assume we have the following sample narrative for “chest pain”:

Clinical document = “Patient is a 49-year-old male with a history of hypertension who presents to the emergency department for evaluation of chest pain. The pain began 3 hours ago gradually at rest, but has since resolved. He describes the pain as sharp and rates it as a 7/10. It is located in the middle of his chest and does not radiate. He denies any fevers, cough, nausea, light-headedness, abdominal pain symptoms.”

Figure 3 – Analyzed SNOMED CT concepts by Amazon Comprehend Medical

Figure 3 – Analyzed SNOMED CT concepts by Amazon Comprehend Medical

Amazon Comprehend Medical identifies various clinical entities and extracts most of the relevant data. In our example “hypertension” is identified as medical condition with traits of “DIAGNOSIS” and “pain” is identified as medical condition with traits of “SYMPTOM” for category of “ANATOMY” equal to chest. It then provides five ordered SNOMED CT concepts and codes referred to in the text. The next step is to extract the response and match the current imaging study attributes like reason for exam, procedure description and body part to the detected entities in our text―linking them to SNOMED CT concepts.

Based on the match we could prioritize the corresponding clinical documents as relevant documents. For each inference, there is also an entity-score representing the underling model’s confidence score of the result. This score can be used to filter results and determine which should be presented to clinicians for confirmation during a complete workflow.

Please note, when using Amazon Comprehend Medical to identify protected health information, please recall that the service provides confidence scores that indicate the level of confidence in the accuracy of the extracted entities. You should evaluate these confidence scores and identify the right confidence threshold for your use case. For specific compliance use cases, we recommend you use additional human review or other methods to confirm the accuracy of extracted PHI.

This will drastically reduce searching through all available clinical documents and narrow it down to just a few. All the other documents, if required, will also be available for any further interpretation in the single patient jacket view.

For more information, refer to the Amazon Comprehend Medical Developer Guide. By default, the API links detected entities to the SNOMED CT US edition. To request support for other editions, contact AWS Support.


Radiology study interpretation and diagnosis is widely accepted to be more effective and accurate when performed with knowledge of, and reference to, the patient’s relevant clinical context and history.

With integrated systems, a radiologist can significantly reduce the time spent searching for patient records during the interpretation process. Sign-off on the report can happen immediately, avoiding unnecessary requests for clarification and approval.

Optimizing the interpretation process decreases the turnaround time for reports and boosts the volume of completed reads. By providing value added services the imaging center can improve radiologist’s productivity.

By leveraging the depth and breadth of AWS, providers can create solutions tailored to improve workflows for clinicians and radiologist needs. We encourage other healthcare providers to start a conversation today with their AWS account team. AWS can help build a solution for you using our ever-growing number of HIPAA-eligible, highly-secure, and reliable services.

To know what AWS can do for you contact an AWS Representative.

Further Reading

Previous Article
Common techniques to detect PHI and PII data using AWS Services
Common techniques to detect PHI and PII data using AWS Services

Identifying sensitive data such as protected health information (PHI) and personally identifiable informati...

Next Article
Top-10 HCLS Announcements from re:Invent
Top-10 HCLS Announcements from re:Invent

re:Invent is an opportunity to dive into innovative technologies and explore new ideas, and for healthcare ...