How to use Text Analytics for health in Azure AI Language

well hello everyone welcome to the Azure AI language learning video series I'm ronak chokshi a product marketer on the Azure AI team here at Microsoft and I'm super excited to share this series where we'll showcase new Cutting Edge natural language processing capabilities based on GPD and other large language models all right let's say you are if you are a healthcare provider or a clinician have you come across situations where you wish you had an automated way to label medical data and extract insights from it using a robust secure and cost effective solution well you've come to the right place in this video Rina Liu a product manager on the Azure ad team leading the text analytics for health product for us will show us how clinicians doctors and hospitals are using the skill to achieve operational efficiencies using Azure language pretty cool over to urina thanks Roanoke we know that text-based data for medical records such as prescriptions and patient reports are often information and detail heavy luckily Azure AI language offers a service called text analytics for health or ta4h which can parse complex text-based medical data and extract keywords and relationships ta4h contains four key functions the named entity recognition function highlights and labels key terms from input text such as diagnosis symptom and treatment The Entity linking function links plain text terms with their unified medical language system metaphosaurus or umls codes the relation extraction function Associates related terms such as treatment type and time of treatment the assertion detection function labels terms with contextual modifiers like certainty conditionality and Association let's take a hypothetical user scenario say we are an insurance company that needs to analyze a prescription to process a claim let's assume that the document has already been converted to a txt file now let's log into Azure AI language Studio to parse these documents in real time you can see that we offer many different language services including summarization translation and more we also offer pre-built and custom analytics services with the pre-built service the machine learning models are already trained and ready to support our four core functions but we can also customize the models with additional data for our hypothetical scenario I'm going to select pre-built text analytics for health foreign I'll start by using this example prescription text it says patient needs to take 100 milligrams of ibuprofen and three milligrams of potassium also needs to take 10 milligrams of Zocor patient needs to take 50 milligrams of ibuprofen and 2 milligrams of Coumadin you can see on the right here that we also have the option to toggle between having the output being returned in fire or fast Healthcare interpretability resources fhir structure which we also have Integrations with and now I'll run the service you can see here that the results are both an original text form and in the format of a Json I'll start by removing the relationship displays for Simplicity as you can see from the input text we have key medical information identified and labeled a hundred milligrams is recognized as a dosage and ibuprofen is recognized as a medication name now if I turn on the relationship toggle we can also see that the analysis covers the functionality of relationship extraction for each of these prescriptions the dose term is correctly linked with the medication name It's associated with we also have Integrations with the unified medical language system metaphasaurus or umls as you can see here we have a umls code for ibuprofen as c0020740 which may be very helpful in logging insurance claims if we are instead looking for the corresponding ATC code for ibuprofen say we're using the anatomical therapeutic chemical classification system then we can select this drop down and see that the ATC code is m01 ae01 we can also parse this information in Json form here on the Json tab we can see again that the first entity term was 100 milligrams and it was labeled as the entity category of dosage with a 99 confidence score we also have the length and the offset listed moving on to the next entity we have ibuprofen recognized in the text and it's been categorized as a medication name with a confidence score of 100 we also see the character offset and character length listed here as well as a list of links showing the various relevant umls codes as you can see this made the task of analyzing your medical text Data fast and easy you can try this out now on language Studio by going to AKA dot Ms slash language Dash Studio to learn more about text analytics for health or any other Azure AI language service go to AKA dot Ms slash Azure Dash language or click on the links in the description of this video