search

BBillot / SynthSeg

Contrast-agnostic segmentation of MRI scans
Apache License 2.0
376 stars 95 forks source link

[question] which modality should I use to get the best results? #96

Closed neuronflow closed 1 month ago

neuronflow commented 1 month ago

I have a brain tumor dataset with multiple modalities.

Usually, T1c has the highest spatial resolution. Which modality promises the best results?

Can I supply multiple modalities, so SynthSeg can exploit all the information available?

hvgazula commented 1 month ago

@BBillot might be able to provide better insights. Nevertheless, synthseg is contrast-agnostic and doesn't have a feature to account for multiple modalities in one go.

BBillot commented 1 month ago

Hi, even though SynthSeg is robust to a large range of contrasts and resolutions, it tends to perform better on T1 scans (that's the modality with best white/grey matter contrast, that's why it's so widely used in brain exams). But overall I'd say use the modality with highest resolution (which is T1c in your case).

Multi-modality is not supported at the moment.

But please be aware that SynthSeg doesn't segment tumours and is has not been trained to be robust to large tumours. Small tumours are okay, but the bigger they get the more likely SynthSeg is to fail. This may bias your results.

Hope this helps, Benjamin