save_volumes

save_volumes(prediction, directory, *, reference, cases=None, suffix='')

A predicted volume as NIfTI, written onto the geometry of the scan it came from.

This is what save_class_masks’s own refusal tells a reader to do, and until now there was nowhere in this package to do it: a PNG carries neither voxel spacing nor orientation, so a volume written as pictures cannot be laid back over its scan. A mask array on its own has the same problem. The affine is the point - every viewer, every registration tool and every volume in millilitres reads position and spacing out of it, and a file carrying the wrong one is either silently misplaced or rejected.

Written as integer labels, int16, because that is what a volume format holds and what the next tool expects to read back. reference is required rather than optional for the same reason: there is no sensible affine to invent.

Refuses an overlapping prediction instead of taking the argmax, exactly as onehot_to_colours does and for the same reason one level along: a label map holds one integer per voxel, so a voxel in both the tumour core and the whole tumour would be written as whichever channel read higher. Those are regions, and save_class_volumes is where they go. Detected from the array rather than from a flag, so a caller holding only a prediction can answer it, and a softmax never trips it.

Parameters

Name	Type	Description	Default
prediction	np.ndarray | Sequence[np.ndarray]	Channels-last probabilities. A list of (d, h, w, c) arrays, which is what predict(space="source") returns, or one such array, or a stacked (n, d, h, w, c).	required
directory	str | Path	Where to write. Created, with parents, if it does not exist.	required
reference	Sequence[str | Path]	The scan each mask was predicted from, in the same order. Its canonical affine is what the mask is written with.	required
cases	Sequence[str] | None	One name per volume, without extension - usually split’s keys. Defaults to mask_0001 and so on. Names that repeat are refused rather than overwritten.	None
suffix	str	Added before the extension, so two models can write into one directory.	''

Returns

Name	Type	Description
	list[Path]	The paths written, in the order the predictions were given.