How to improve fill percentage for Ellipsoid Factor

I have run the ellipsoid factor (EF) analysis on a rat tibia trabecular VOI created in CTan.

The only pre-processing I have run in ImageJ is auto-thresholding.

The log tells me that the fill percentage is ~72% of foreground volume filled with ellipsoids.

I have read another post on this forum from Alessandro Felder that says above 90% fill percentage is what I should be aiming for, to ensure I properly represent the sample.

Does anyone know the best pre-processing stages for improving the fill percentage?

In the image, you can see the binarized dataset alongside the EF diagram for the same slice. Inside the blue box, you can see the an example structure which has been somewhat diminished by running the EF plugin. Do structural changes like this result in a worse fill percentage?

Best regards,