Quickstart: Use the Radiology Insights model
This quickstart provides an overview on how to use the Radiology Insights.
Prerequisites
To use the Radiology Insights (Preview) model, you must have an Azure AI Health Insights service created.
If you have no Azure AI Health Insights service, see Deploy Azure AI Health Insights using the Azure portal.
Once deployment is complete, you use the Azure portal to navigate to the newly created Azure AI Health Insights service to see the details, including your Service URL.
The Service URL to access your service is: https://YOUR-NAME.cognitiveservices.azure.com.
Example request and results
To send an API request, you need the endpoint and key of your Azure AI Health Insights service.
You can also find a full view of the request parameters here.
Important
Prediction is performed upon receipt of the API request and the results will be returned asynchronously. The API results are available for 24 hours from the time the request was ingested, and is indicated in the response. After this time period, the results are purged and are no longer available for retrieval.
Example request
Note
The examples below are based on API version: 2024-04-01. There might be changes between API versions. For a specific API version, please use the reference to the REST API to see full description.
Starting with a request that contains a case
You can use the data from this example, to test your first request to the Radiology Insights model.
Definition {jobid}
- unique identifier
- maximum 36 characters
- no spaces
PUT
https://{cognitive-services-account-endpoint}/health-insights/radiology-insights/jobs/{jobid}?api-version=2024-04-01
Content-Type: application/json
Ocp-Apim-Subscription-Key: {cognitive-services-account-key}
{
"jobData": {
"configuration": {
"inferenceOptions": {
"followupRecommendationOptions": {
"includeRecommendationsWithNoSpecifiedModality": false,
"includeRecommendationsInReferences": false,
"provideFocusedSentenceEvidence": false
},
"findingOptions": {
"provideFocusedSentenceEvidence": false
}
},
"inferenceTypes": ["lateralityDiscrepancy"],
"locale": "en-US",
"verbose": false,
"includeEvidence": false
},
"patients": [
{
"id": "111111",
"details": {
"sex": "female",
"birthDate" : "1986-07-01T21:00:00+00:00",
"clinicalInfo": [
{
"resourceType": "Observation",
"status": "unknown",
"code": {
"coding": [
{
"system": "http://www.nlm.nih.gov/research/umls",
"code": "C0018802",
"display": "MalignantNeoplasms"
}
]
},
"valueBoolean": "true"
}
]
},
"encounters": [
{
"id": "encounterid1",
"period": {
"start": "2021-8-28T00:00:00",
"end": "2021-8-28T00:00:00"
},
"class": "inpatient"
}
],
"patientDocuments": [
{
"type": "note",
"clinicalType": "radiologyReport",
"id": "docid1",
"language": "en",
"authors": [
{
"id": "authorid1",
"fullName": "authorname1"
}
],
"specialtyType": "radiology",
"createdAt": "2021-8-28T00:00:00",
"administrativeMetadata": {
"orderedProcedures": [
{
"code": {
"coding": [
{
"system": "Https://loinc.org",
"code": "26688-1",
"display": "US BREAST - LEFT LIMITED"
}
]
},
"description": "US BREAST - LEFT LIMITED"
}
],
"encounterId": "encounterid1"
},
"content": {
"sourceType": "inline",
"value" : "Exam: US LT BREAST TARGETED\r\n\r\nTechnique: Targeted imaging of the right breast is performed.\r\n\r\nFindings:\r\n\r\nTargeted imaging of the left breast is performed from the 6:00 to the 9:00 position. \r\n\r\nAt the 6:00 position, 5 cm from the nipple, there is a 3 x 2 x 4 mm minimally hypoechoic mass with a peripheral calcification. This may correspond to the mammographic finding. No other cystic or solid masses visualized.\r\n"
}
}
]
}
]
}
}
You can also find a full view of the request parameters here.
Evaluating a response that contains a case
You get the status of the job by sending a request to the Radiology Insights model by adding the job ID from the initial request in the URL.
Example code snippet:
GET
https://{cognitive-services-account-endpoint}/health-insights/radiology-insights/jobs/{jobid}?api-version=2024-04-01
Ocp-Apim-Subscription-Key: {cognitive-services-account-key}
{
"result": {
"patientResults": [
{
"patientId": "11111",
"inferences": [
{
"kind": "lateralityDiscrepancy",
"lateralityIndication": {
"coding": [
{
"system": "http://snomed.info/sct",
"code": "24028007",
"display": "RIGHT (QUALIFIER VALUE)"
}
]
},
"discrepancyType": "orderLateralityMismatch"
}
]
}
]
},
"id": "jobid",
"createdDateTime": "2023-12-18T12:25:37.8942771Z",
"expirationDateTime": "2023-12-18T12:42:17.8942771Z",
"lastUpdateDateTime": "2023-12-18T12:25:49.7221986Z",
"status": "succeeded"
}
You can also find a full view of the request parameters here.
Data limits
Limit, Value
- Maximum # patients per request, 1
- Maximum # patientdocuments per request, 1
- Maximum # encounters per request, 1
- Maximum # characters per patient, 50,000 for data[i].content.value all combined
Request validation
Every request contains required and optional fields that should be provided to the Radiology Insights model. When you're sending data to the model, make sure that you take the following properties into account:
Within a request:
- patients should be set
- patients should contain one entry
- ID in patients entry should be set
Within configuration: If set, configuration locale should be one of the following values (case-insensitive):
- en-CA
- en-US
- en-AU
- en-DE
- en-IE
- en-NZ
- en-GB
Within patients:
- should contain one patientDocument entry
- ID in patientDocument should be set
- if encounters and/or info are used, ID should be set
For the patientDocuments within a patient:
- createdDateTime (serviceDate) should be set
- Patient Document language should be EN (case-insensitive)
- documentType should be set to note (case-insensitive)
- Patient Document clinicalType should be set to radiologyReport or pathologyReport (case-insensitive, in one word)
- Patient Document specialtyType should be radiology or pathology
- If set, orderedProcedures in administrativeMetadata should contain code -with code and display- and description
- Document content shouldn't be blank/empty/null
Optional: sex and birthDate are optional fields.
"patientDocuments" : [ {
"type" : "note",
"clinicalType" : "radiologyReport",
"id" : "docid1",
"language" : "en",
"authors" : [ {
"id" : "authorid1",
"name" : "authorname1"
} ],
"specialtyType" : "radiology",
"createdDateTime" : "2021-8-28T00:00:00",
"administrativeMetadata" : {
"orderedProcedures" : [ {
"code" : {
"coding" : [ {
"system" : "Https://loinc.org",
"code" : "41806-1",
"display" : "CT ABDOMEN"
} ]
},
"description" : "CT ABDOMEN"
} ],
"encounterId" : "encounterid1"
},
"content" : {
"sourceType" : "inline",
"value" : "CT ABDOMEN AND PELVIS\n\nProvided history: \n78 years old Female\nAbnormal weight loss\n\nTechnique: Routine protocol helical CT of the abdomen and pelvis were performed after the injection of intravenous nonionic iodinated contrast. Axial, Sagittal and coronal 2-D reformats were obtained. Oral contrast was also administered.\n\nFindings:\nLimited evaluation of the included lung bases demonstrates no evidence of abnormality. \n\nGallbladder is absent. "
}
} ]
Next steps
To get better insights into the request and responses, you can read more on following pages: