How emtelliPro works

Entity Linking This is the ability of emtelliPro to perform what people think of as ‘medical coding’ – it is taking words in the unstructured text, and assigning them a concept ID from a medical ontology to explain what their meaning is. In the sentence below, the following SNOMED-CT concepts are identified:

• CT Scan – 77477000 | Computed tomography (procedure) |
• appendicitis – 74400008 | Appendicitis (disorder) |
• adjacent – 18769003 | Juxta-posed (qualifier value) |
• abscess – 128477000 | Abscess (disorder) |

Polarity Detection – understanding whether a medical concept is present or absent. In the sentence below, the cross-hatched background indicates that the terms `chest pain`, `shortness of breath`, etc are all negated – the patient did not have these symptoms.

Uncertainty Detection– knowing whether or not the author was certain or speculative about a diagnosis. In the sentence below, lupus is tagged with a `U` to indicate it was an uncertain diagnosis.

Disambiguation– understanding the context of a phrase or statement that may have more than one meaning. In the sentences below, we see 3 uses of ‘PT’ – meaning `prothrombin time`, `patient`, and `physiotherapy service` (we are also seeing a measurement relation between `prothrombin time` and `15.4`):

Experiencer Detection – determining if the symptom or condition was the patient or someone else (e.g. family member) or the patient themselves

Measurement Detection– finding measurements in the unstructured text. The following sentence is a list of results from various blood tests; it can be seen below how emtelliPro identifies all the tests, and associates the test result measurement with the test:

Temporality Relations – determining when a procedure occurred or condition was diagnosed, or if an event was historical. In the first example below, the concepts `echocardiogram`, `ejection fraction`, and `mild mitral regurgitation` are connected to the date they were identified on (`12/18/1980`), and identifies the measurement relation of `55% to 60%` with the `ejection fraction` concept. In the second example, the historical events of `rectal cancer`, and `chemo and radiation` are linked to the historical qualifiers `history of` and `status post`.

Medication Identification– detecting the drug name, dose, frequency, route and other prescription information.

Follow Up Detection– finding specific recommendations for additional imaging or clinical action.