We implemented the following chest x-ray algorithm.
docker-compose up -d --build
This will start two docker services
- algorithm-inference-api: REST API and Scheduler service to save incoming requests to database and process them in a sequential manner.
- algorithm-mysql: Stores all the incoming requests and results
You can change the ports and environment vairables by changing the docker-compose.yml file.
Once the services are up and running, you can now integrate the algorithm onto your CARPL account. You can refer to this guide for the same.
You can use the existing code base in this repository. Requirements:
- Update the
predictionfunction in the filealgorithm_framework/main.pyand your prediction function logic. - This sample algorithm accepts a DICOM file[.dcm format]. To handle other file formats, add a preprocessing step in the
/api/uploadfunction. - Outputs of the two inferencing APIs should adhere to CARPL accepted formats. Please refer to the samples below.
Sample
/api/resultsresponse:
{
"findings": [
{
"name": "Finding_A",
"probability": "19.565606117248528"
},
{
"name": "Finding_B",
"probability": "34.508439898490906"
},
],
"status": "Processed",
"job_id": "<UNIQUE_JOB_ID>"
}
Sample /api/bbox response:
{
"rois": [
{
'type': 'Freehand',
'StudyInstanceUID': 'XXXXXXXXX',
'SeriesInstanceUID': 'XXXXXXXXXXXX',
'SOPInstanceUID': 'XXXXXXXXXX',
'points': [[point_1_x, point_1_y],[point_2_x, point_2_y],[point_3_x, point_3_y], ... ,[point_n_x, point_n_y]],
'finding_name': 'XXXX'
},
{
'type': 'Rectangle',
'StudyInstanceUID': 'XXXXXXXXX',
'SeriesInstanceUID': 'XXXXXXXXXXXX',
'SOPInstanceUID': 'XXXXXXXXXX',
'points': [[top_left_x, top_left_y], [bottom_right_x, bottom_right_y]],
'finding_name': 'XXXX'
}
],
"status": "Processed",
"job_id": "<UNIQUE_JOB_ID>"
}
NOTE: If you have any concerns running this repository, please write to our team at [email protected].