A new vendor partnership aims to bring real-world evidence related to medical images to Amazon Web Service’s AWS Data Exchange.
The collaboration, between Life Image and Graticule is expected to make available a de-identified patient-level data set with information from natural language processing derived from Life Image’s data.
The goal of the initiative is to provide a cost-effective patient-level snapshot to biopharma data science teams to support discussions about the data, leading to sourcing broad curated sets or building artificial intelligence models.
Executives from Life Image—a medical evidence network for clinical and imaging data—and Graticule, a real-world data firm, say they are working together to make the data set available through data licensing subscriptions on AWS Data Exchange, a new service that enables customers of Amazon Web Services (AWS) to securely to and use third-party data in the cloud.
Graticule has published the Graticule Life Image Machine Parsed Set (GLIMPS) as a de-identified patient level data set, including a summary of features generated from NLP processing from Life Image data. Life Image’s HIPAA-compliant repository of diagnostic images provides insights from a broad set of geographically diverse providers in the U.S.
Because much of the key data in imaging is stored in unstructured reports or within images, it is difficult to construct queries to identify studies with features of interest to solve researcher questions. GLIMPS provides a structured view of the patient-level medical information by providing coded values using open vocabularies such as ICD9 or SNOMED to execute feasibility analysis.
“Diagnostic images and radiology reports contain interpretations and other information that provide rich diagnostic and outcomes data but have historically been extremely difficult to access and aggregate at scale,” said Dan Housman, chief technology officer and co-founder of Graticule. “AWS Data Exchange provides a sea change opportunity that allows us to distribute our data to provide a transparent view of available images and tools to deliver deeper data insights on demand.”
The GLIMPS data model includes key fields from DICOM headers and a machine-parsed mapping created using multiple medical NLP tools. Custom extraction approaches provide an additional deeper layer of clinical context using measurable values, such as ejection fraction within echocardiograms.
The data model was built to be compatible with the OHDSI (Observational Health Data Sciences and Informatics) Common Data Model to rapidly integrate into Real World Data tools used by Life Sciences companies and regulators.
Executives of the company say uses for the data include:
- Longitudinal data to identify markers of disease prior to diagnosis and progression during treatment.
- Supply machine learning teams with “control” data from Life Image vs. “study” data from a clinical trial or internal data source.
- Research into identification of novel imaging biomarkers for a rare or newly identified disease subtype.
- Validate AI models on real-world data that have been built on small or narrow data sets.
- Support for real-world evidence teams with a novel source of data to support new approaches or techniques.
“As one of the largest evidence-based networks focused on multivariable medical data, Life Image has been solving for those technical challenges, which is why we are able offer Real World Imaging,” says Matthew A. Michela, president and CEO of Life Image. “Launching GLIMPS in AWS Data Exchange will allow biopharma and AI companies to accelerate development, post-market safety, effectiveness, validation, and label expansion.”