Dynamic Time Warping Nilean

Hi folks,
I have read the following interesting paper, using Dynamic Time Warping as a way of improving the measure of time series similarity between subjects:

Is this possible to do within Nilearn?

If anyone has any experience in this please do reach out.

Thanks,
Chris

No, there is no such method in Nilearn.
What would be the use case of such a method ?
Best,

Hi Bertrand,
I have been researching this more in depth, and it seems to be another connectivity parameterization method. It is best explained here:
https://ieeexplore.ieee.org/document/7760247

Using DTW would account for small time delay differences between activated regions, therefore creating a more accurate picture of the correlation between regions in contrast to the standard pearson correlation.

If you have any thoughts on this I would love to hear them.

Thanks,
Chris

This is interesting, even though I tend to be cautious with short, single-experiment paper.
Can you reproduce on one of the resting-state datasets of Nilearn ?
Best,

Hi Bertrand,
I have continued working on this and I have replicated the experiment and classification from the following example on the Nilearn site: https://nilearn.github.io/auto_examples/03_connectivity/plot_group_level_connectivity.html#sphx-glr-auto-examples-03-connectivity-plot-group-level-connectivity-py

It achieves a better accuracy than standard correlation with 88%.

image

Please find the Jupyter Notebook on github here: https://github.com/FL33TW00D/DTW
I did not implement it in the ConnectivityMeasure object yet as this was a short experiment.

Let me know what you think.

Thanks,
Chris

I’m afraid that the gain is quite limited. I would suggest to make it a side project that is compatible with Nilearn, and consider it if we observe systematic gains in many settings and support from the community.

Hi Bertand,
I appreciate you looking at it and I will continue to refine the method.

Many Thanks,
Chris