Questions about rat brain support (beta) and alignment performance

[ARTUR551]

Hi, thanks for making DeepSlice available, it’s very useful.

I had a few questions specifically about the rat brain support:

The README mentions that rat alignment (Waxholm atlas) is still in beta. From your experience, how reliable is the rat model currently?

Have you (or others) successfully aligned full rat brains or typical experimental datasets with good accuracy? Any examples or advice would be very helpful for me.

In my own tests with rat data, the results are quite inconsistent, is this expected given the current training/data limitations?

Are there known constraints (for example, staining type, section quality, or dataset size, etc.) that particularly affect rat performance?

I hope you still have plans to continue improving the rat model in the upcoming versions...

For context, I’m following the recommended workflow (coronal sections, consistent dataset, etc.), but still seeing weaker performance than expected. The online tool does not help either...

Thanks in advance for any clarification! And of course for the great tool.

A.

[PolarBean]

Hey Arthur,

Yes the DeepSlice rat model is significantly weaker than the mouse model. This is why we havent yet published an article on it. We have paused work on the rat model for the moment as the main limitation is the amount of training data available.

I havent compared different stains to be able to confidently draw any useful conclusions.

Sorry I dont have better news!

All the best,

Harry

[PolarBean]

were you using the web version or Python version. I think i discovered an issue with the web version that has now been fixed.