inline-algorithm-sdk

SDK Documentation: https://developers.pramana.ai/inline-algorithms/sdk/documentation

API Documentation: https://developers.pramana.ai/inline-algorithms/documentation

Source Code: https://github.com/lumenbiomics/inline-algorithm-sdk

---

Requirements

• Needs Python3.10 or higher

---

Installation

$ python3 -m venv venv
$ source venv/bin/activate
$ pip install inline_algorithm@git+https://github.com/lumenbiomics/inline-algorithm-sdk.git

---

Example to run locally

• Create a file main.py with:

from inline_algorithm.inline_algo_queue_processor import InlineAlgoQueueProcessor

class TestChild(InlineAlgoQueueProcessor):
    def __init__(self, port, host, docker_mode=True):
        super().__init__(port, host, docker_mode)

    #optional
    def on_server_start(self):
        print('loading model')
        return
    #optional
    def on_server_end(self):
        print('freeing memory')
        return
    
    #update this function with your machine learning model detection/segmentation helper
    def process(self, message):
        """
        Processes a message of type ScanOngoing.

        Parameters:
        -----------
        message : ScanOngoing
            An instance of the ScanOngoing class.

        The ScanOngoing class is defined as follows:
        -------------------------------------------
        class ScanOngoing(BaseModel):
            slide_name: str
                The name of the slide being scanned.
            tile_name: str
                The name of the tile within the slide.
            tile_image_path: str
                The file path to the image of the tile.
            row_idx: int
                The row index of the tile in the grid.
            col_idx: int
                The column index of the tile in the grid.

        Returns:
        --------
        List[DetectionArray]
            A list containing the processed data. Example:
                [
                    {"bbox": [0, 0, 1, 1], "confidence": 0.9, "class": "tumor"},
                    {"bbox": [123, 321, 278, 456], "confidence": 0.6, "class": "stroma"},
                ]
            The bounding box (bbox) value contains information about model detections in the
            form of [x1, y1, x2, y2].

            OR

        List[List] (to be deprecated in the future)
            A two-dimensional list with the processed data. Ex - [[0, 0, 0.9, 1.2, 0.8, "tumor"], [123, 321, 200, 400, 0.6, "stroma"]].
            Each sub array contains information about bounding boxes like [x1, y1, x2, y2, confidence, class strings].
        """
        return 
    

if __name__== '__main__':
    obj = TestChild(8000, 'localhost', docker_mode=False)
    obj.run()

• Run the server: python main.py
• This will run the Fast API server and will be able to handle all the API requests
• You can test this example with the Pramana API call simulator which will simulate the Pramana scanner's scanning pipeline. See this section

---

Test with the simulator

• Clone this repository and follow the instructions in this README.md to setup the simulator

$ git clone https://github.com/lumenbiomics/inline-algorithm-sdk

• The simulator will emulate the Pramana scanner and simulate making API calls to port 8000 (as specified in the above example).

---

Dockerize the Inline Algorithm

• Create a file called Dockerfile with these contents

# Use the official Python image from Docker Hub
FROM python:3.9-slim

ENV PYTHONUNBUFFERED=1
ENV DEBIAN_FRONTEND noninteractive
WORKDIR /app

RUN apt-get update

#Add any other files you need
COPY test.py  .
#COPY <your-requirements.txt-file> if any

# Install dependencies
RUN pip install inline_algorithm@git+https://github.com/lumenbiomics/inline-algorithm-sdk.git
# Install any other dependencies
#RUN pip install -r <your-requirements.txt-file> if any

# Expose port 8000 to the outside world
EXPOSE 8000

# Command to run the application
CMD ["python3", "main.py"]

• Run the following command to build your Docker image:

$ docker build -t <image name>:<tag> .

• Create a file called docker-compose.yml with these contents

version: '3.9'
x-pramana-ui-name: Pramana Simulator
services:
  model:
    image: '<image name>:<tag>'
    volumes:
      - '<local basepath from your machine>/pramana_api_call_simulator/data:/data/acquired_data'
    ports:
      - '8000:8000'
    extra_hosts:
      - 'host.docker.internal:host-gateway'

Note: For the docker-compose.yml requirements that we need for the Inline Algorithm to run on the Pramana Scanner, please refer to the official API documentation here. In this example above, the volume mount and ports are specifically set to work with the simulator and NOT the Pramana Scanner.

• Turn on the container using this command : docker compose -f <path to the docker compose file> up