log transform in masking fails for some (spinal) datasets

maybe make this optional or use otsu

Locate the masking step that applies the log transform and reproduce the failure on the affected spinal datasets. Read how masking inputs are loaded and how the current transform is selected, then compare the observed behavior with the issue's optional-transform or Otsu suggestions. Done means masking works reliably for those datasets with a documented, testable behavior.

Written by the indexing model from the issue text.