Discrepancy between modalities classified - paper vs code

[deepakri201]

Hi,

Thank you for this great toolbox. I've been trying out your pre-trained model, and have noticed a possible discrepancy between the modalities classified.

In your paper, I see that there are 12 modalities classified:

B-Value Volumes, T2-weighted (T2w), T1-weighted (T1w), Apparent Diffusivity Coefficient (ADC), Field Map, Proton Density (PD), Fluid-Attenuated Inversion Recov- ery (FLAIR), functional MRI (fMRI), Fractional Anisotropy (FA), T2*-weighted (T2star), Short Tau Inversion Recovery(STIR), exponential ADC (eADC).

However, in your code https://github.com/BRAINSia/dcm-classifier/blob/main/src/dcm_classifier/image_type_inference.py I see the following from lines 33-44, which shows 10 modalities:

imagetype_to_integer_mapping = {
    "t1w": 0,
    "gret2star": 1,
    "t2w": 2,
    "flair": 3,
    "b0": 4,
    "tracew": 5,
    "adc": 6,
    "fa": 7,
    "eadc": 8,
    "dwig": 9,
}

Thank you,

Deepa

[mbrzus]

Hi Deepa,

You correctly spotted the difference! Initially, we started with 12 classes but over time narrowed it down to 10 because:

• We didn't have enough robust data for some of the original modalities.
• It didn't match up with what we needed for our current projects.

We continue our development, so the list of supported modalities might evolve. In the near future, we don't have plans to expand the supported modalities, but if you need to adapt the model for other data, we've got a tutorial on how to retrain it.

Thanks for using our tool, and let us know if you have any other questions!

Best,
Michal