AWS adds ontology linking to Comprehend Medical natural language processing service

Comprehend Medical uses machine learning to model topics, detect language, conduct sentiment analysis and extract phrases from unstructured medical texts.

Amazon's natural language processing service for the healthcare industry, Comprehend Medical, is now capable of linking information to medical ontologies. 

Comprehend Medical uses machine learning to model topics, detect language, conduct sentiment analysis and extract phrases from unstructured medical texts. Comprehend Medical is also designed to understand relationships between things like dosage and conditions.

With the new ontology-linking feature, Amazon said clinicians will be able to detect medication and medical information in unstructured clinical text, and link that information to the ICD-10-CM and RxNorm medical ontologies. In addition to improving medical outcomes and care, the feature is meant to help clinicians reduce the cost, time and effort it takes to process large amounts of unstructured medical text, Amazon said.

Also: The real future of healthcare is cultural change, not just AI and other technology 

Historically, hospitals, pharmacies and other medical service organizations have used a variety of naming concepts in their computer systems. The goal of RxNorm and other medical ontologies is to enable efficient communication of information between these computer systems. Comprehend Medical fits in as a system to link detected entities, like symptoms and diagnoses, from medical records to specific concepts in ontologies.

"Using Amazon Comprehend Medical ICD-10-CM and RXNorm Ontology Linking APIs, developers can quickly and accurately extract codes (e.g. 'R51' as the ICD-10-CM code for headache) from a variety of data sources, such as doctor's notes or patient health records," the company wrote in an Amazon Web Services blog post. "Our deep learning approach to ontology linking provides much higher accuracy than existing rules-based systems by understanding the context each entity is found in." 

RELATED: