how to handle different anatomical region specifically

1. I noticed that the task description mentions that it's possible to use different models for different organs. However, the submission interface does not seem to distinguish between organs. In the docker algorithm template provided by synthRAD2023, I also didn’t find any relevant handling for this. Does this mean I need to determine the organ type from each case by reading the filename and then select the corresponding model accordingly?

2. Can I package my own framework into a Docker container instead of integrating it into the provided docker algorithm template? (Because my framework differs significantly from the provided template in terms of data loading and preprocessing.) I tried to do so using the information from the synthRAD2023 template and successfully passed the test.sh. However, when I submitted it on the synthRAD2023 website, it still failed. According to the debugging instructions provided on the website, I couldn’t find the corresponding log file.

Dear intelland2024,

thank you for raising the points.

1) The algorithms will have access to input with MRI or CBCT, mask and a region.json. In the JSON file, you can retrieve the information regarding the regions with as possible values: "Head and Neck", "Abdomen" or "Thorax". Here just an example of the locations.

"outputs": [ { "type": "metaio_image", "filename": "/output/images/synthetic-ct/1BAxxx.nii.gz" } ], "inputs": [ { "type": "metaio_image", "filename": "/input/images/mri/1BAxxx.nii.gz" }, { "type": "metaio_image", "filename": "/input/images/body/1BAxxx.nii.gz" }, { "type": "String", "filename": "/input/region.json" } ]

2) Yes, you can as long as you set up the algorithm to work on a case base (so one patient with MRI or CBCT, one mask and one region) per time and you read in the location that we will specify. During SynthRAD2023, access to the log file on Grand Challenge was not possible, now it should be feasible, we are checking which rights as we speak. I foresee that in the preliminary phase the log will be fully accessible to allow debugging, while in the test phase participants will not have access to it. On June, 1 we will open up a new template for the algorithm at https://github.com/SynthRAD2025/algorithm-template with all the necessary information.

I would like to confirm whether each call to docker run processes only a single case at a time. Specifically:

1. Will each execution of the algorithm container receive exactly one MRI or CBCT image, one mask, and one region.json file corresponding to a single anatomical region?

2. Or should the algorithm be designed to handle multiple cases within a single docker run?

This clarification is important for me to determine how to structure the preprocessing and model selection logic within my Docker container.

Yes, each algorithm processed one job at a time.

Please see the general documentation from GC: https://grand-challenge.org/documentation/flexible-inputs-and-outputs/.

In less than 10 days, the algorithm template will be opened with some suggestions in it, so stay tuned!