Overview

Annotator for Clinical Data is designed to find medical concepts, medical codes, and contextual information in unstructured text. It provides turnkey annotators as well as highly customizable annotators that you can tune specifically for your application needs. The Unified Medical Language System (UMLS is the primary source for concepts and medical codes. Annotator for Clinical Data can also work with user-provided ontologies beyond UMLS.

To illustrate the basic function of Annotator for Clinical Data, let’s look at a simple example. Imagine that we have the following small snippet of text:

The resulting concept over ulceration will contain medical codes along with contextual information about the concept (note in this example, the concept is negated):

User configuration data: User configuration data is stored in Cloud Object Storage. This data is isolated by tenant at the service layer.

User unstructured text: Requested analysis of unstructured text is processed in-memory. The input text is not persisted. Requests are isolated at the service layer.

User configuration data: User configuration data is stored in Cloud Object Storage. This data is isolated at the service layer.

User unstructured text: Requested analysis of unstructured text is processed in-memory. The input text is not persisted. Requests are isolated at the Kubernetes pod layer.

Public multi-tenant instances: The service is accessed via public endpoints. The service will access all dependencies via private endpoints. In these instances the control plane of the service is shared across tenants.

Dedicated instances: The service can be accessed via public and/or private endpoints. Review Public and private network endpoints for details. The service will access all dependencies via private endpoints. In these instances the control plane of the service is isolated in separate pods for each tenant.

The following annotators are available for detecting and coding medical concepts within unstructured data.

Attributes are higher order concepts composed of multiple pieces of information found in a document. An example of this might be understanding if a patient is overweight or not. Given some example text, we would like to know if the patient is overweight or not.

You can create custom concept values to extract and normalize the patient’s height and weight. With that information, you can create inference rules in Annotator for Clinical Data to combine that information into a single attribute that we’ll call NORMAL_WEIGHT. Custom attributes like this are a powerful way to distill unstructured text into actionable insights.

For more information, see Attributes.

The concept annotator finds UMLS or custom concepts in unstructured text.

For more information, see Concepts.

The concept value annotator creates composite attributes resulting from a medical concept and an associated value. It supports scalar values as well as value ranges.

In this example, the combination of height with 6 feet is an example of how concept values work.

For more information, see Concept Value.

Contextual annotators use the surrounding context of the document to provide a deeper understanding of concepts.

Identifies spans of text with an implied negative meaning. For example: There were no signs of ulceration.

For more information, see Negation.

Identifies spans of text are the object of a hypothetical statement. For example: We discussed the pros and cons of chemotherapy.

For more information, see Hypothetical.

Determines the validity of UMLS concepts detected in a document.

For more information, see Disambiguation.

Medically aware spell checker that can be integrated into an API call.

For more information, see Spell Check.

Annotator for Clinical Data provides a set of prebuilt annotators targeted at specific medical domains.

Clinical insights are a collection of models and cartridge configuration that provide contextual information about key clinical attributes (medication, diagnosis, and procedure) for patient-centric clinical notes.

For more information, see Clinical Insights.