test_morph.py

# Authors: The MNE-Python contributors.
# License: BSD-3-Clause
# Copyright the MNE-Python contributors.

from inspect import signature

import numpy as np
import pytest
from numpy.testing import assert_allclose, assert_array_equal, assert_array_less
from scipy.sparse import csr_array
from scipy.sparse import eye as speye
from scipy.spatial.distance import cdist

import mne
from mne import (
    SourceEstimate,
    SourceMorph,
    VectorSourceEstimate,
    VolSourceEstimate,
    VolVectorSourceEstimate,
    compute_source_morph,
    get_volume_labels_from_aseg,
    grade_to_vertices,
    make_ad_hoc_cov,
    make_forward_solution,
    make_sphere_model,
    read_evokeds,
    read_forward_solution,
    read_freesurfer_lut,
    read_source_estimate,
    read_source_morph,
    setup_volume_source_space,
)
from mne._freesurfer import _get_atlas_values, _get_mri_info_data
from mne.datasets import testing
from mne.fixes import _get_img_fdata
from mne.minimum_norm import apply_inverse, make_inverse_operator, read_inverse_operator
from mne.source_space._source_space import _add_interpolator, _grid_interp
from mne.transforms import quat_to_rot
from mne.utils import _record_warnings, catch_logging

# Setup paths

data_path = testing.data_path(download=False)
sample_dir = data_path / "MEG" / "sample"
subjects_dir = data_path / "subjects"
fname_evoked = sample_dir / "sample_audvis-ave.fif"
fname_trans = sample_dir / "sample_audvis_trunc-trans.fif"
fname_inv_vol = sample_dir / "sample_audvis_trunc-meg-vol-7-meg-inv.fif"
fname_fwd_vol = sample_dir / "sample_audvis_trunc-meg-vol-7-fwd.fif"
fname_vol_w = sample_dir / "sample_audvis_trunc-grad-vol-7-fwd-sensmap-vol.w"
fname_inv_surf = sample_dir / "sample_audvis_trunc-meg-eeg-oct-6-meg-inv.fif"
fname_aseg = subjects_dir / "sample" / "mri" / "aseg.mgz"
fname_fmorph = data_path / "MEG" / "sample" / "fsaverage_audvis_trunc-meg"
fname_smorph = sample_dir / "sample_audvis_trunc-meg"
fname_t1 = subjects_dir / "sample" / "mri" / "T1.mgz"
fname_vol = subjects_dir / "sample" / "bem" / "sample-volume-7mm-src.fif"
fname_aseg = subjects_dir / "sample" / "mri" / "aseg.mgz"
fname_fs_vol = subjects_dir / "fsaverage" / "bem" / "fsaverage-vol7-nointerp-src.fif.gz"
fname_aseg_fs = subjects_dir / "fsaverage" / "mri" / "aseg.mgz"
fname_stc = sample_dir / "fsaverage_audvis_trunc-meg"

pytest.importorskip("nibabel")


def _real_vec_stc():
    inv = read_inverse_operator(fname_inv_surf)
    evoked = read_evokeds(fname_evoked, baseline=(None, 0))[0].crop(0, 0.01)
    return apply_inverse(evoked, inv, pick_ori="vector")


def test_sourcemorph_consistency():
    """Test SourceMorph class consistency."""
    assert (
        list(signature(SourceMorph.__init__).parameters)[1:-1]
        == mne.morph._SOURCE_MORPH_ATTRIBUTES
    )


@testing.requires_testing_data
def test_sparse_morph():
    """Test sparse morphing."""
    rng = np.random.RandomState(0)
    vertices_fs = [
        np.sort(rng.permutation(np.arange(10242))[:4]),
        np.sort(rng.permutation(np.arange(10242))[:6]),
    ]
    data = rng.randn(10, 1)
    stc_fs = SourceEstimate(data, vertices_fs, 1, 1, "fsaverage")
    spheres_fs = [
        mne.read_surface(subjects_dir / "fsaverage" / "surf" / f"{hemi}.sphere.reg")[0]
        for hemi in ("lh", "rh")
    ]
    spheres_sample = [
        mne.read_surface(subjects_dir / "sample" / "surf" / f"{hemi}.sphere.reg")[0]
        for hemi in ("lh", "rh")
    ]
    morph_fs_sample = compute_source_morph(
        stc_fs,
        "fsaverage",
        "sample",
        sparse=True,
        spacing=None,
        subjects_dir=subjects_dir,
    )
    stc_sample = morph_fs_sample.apply(stc_fs)
    offset = 0
    orders = list()
    for v1, s1, v2, s2 in zip(
        stc_fs.vertices, spheres_fs, stc_sample.vertices, spheres_sample
    ):
        dists = cdist(s1[v1], s2[v2])
        order = np.argmin(dists, axis=-1)
        assert_array_less(dists[np.arange(len(order)), order], 1.5)  # mm
        orders.append(order + offset)
        offset += len(order)
    assert_allclose(stc_fs.data, stc_sample.data[np.concatenate(orders)])
    # Return
    morph_sample_fs = compute_source_morph(
        stc_sample,
        "sample",
        "fsaverage",
        sparse=True,
        spacing=None,
        subjects_dir=subjects_dir,
    )
    stc_fs_return = morph_sample_fs.apply(stc_sample)
    offset = 0
    orders = list()
    for v1, s, v2 in zip(stc_fs.vertices, spheres_fs, stc_fs_return.vertices):
        dists = cdist(s[v1], s[v2])
        order = np.argmin(dists, axis=-1)
        assert_array_less(dists[np.arange(len(order)), order], 1.5)  # mm
        orders.append(order + offset)
        offset += len(order)
    assert_allclose(stc_fs.data, stc_fs_return.data[np.concatenate(orders)])


@testing.requires_testing_data
def test_xhemi_morph():
    """Test cross-hemisphere morphing."""
    stc = read_source_estimate(fname_stc, subject="sample")
    # smooth 1 for speed where possible
    smooth = 4
    spacing = 4
    n_grade_verts = 2562
    stc = compute_source_morph(
        stc,
        "sample",
        "fsaverage_sym",
        smooth=smooth,
        warn=False,
        spacing=spacing,
        subjects_dir=subjects_dir,
    ).apply(stc)
    morph = compute_source_morph(
        stc,
        "fsaverage_sym",
        "fsaverage_sym",
        smooth=1,
        xhemi=True,
        warn=False,
        spacing=[stc.vertices[0], []],
        subjects_dir=subjects_dir,
    )
    stc_xhemi = morph.apply(stc)
    assert stc_xhemi.data.shape[0] == n_grade_verts
    assert stc_xhemi.rh_data.shape[0] == 0
    assert len(stc_xhemi.vertices[1]) == 0
    assert stc_xhemi.lh_data.shape[0] == n_grade_verts
    assert len(stc_xhemi.vertices[0]) == n_grade_verts
    # complete reversal mapping
    morph = compute_source_morph(
        stc,
        "fsaverage_sym",
        "fsaverage_sym",
        smooth=smooth,
        xhemi=True,
        warn=False,
        spacing=stc.vertices,
        subjects_dir=subjects_dir,
    )
    mm = morph.morph_mat
    assert mm.shape == (n_grade_verts * 2,) * 2
    assert mm.size > n_grade_verts * 2
    assert mm[:n_grade_verts, :n_grade_verts].size == 0  # L to L
    assert mm[n_grade_verts:, n_grade_verts:].size == 0  # R to L
    assert mm[n_grade_verts:, :n_grade_verts].size > n_grade_verts  # L to R
    assert mm[:n_grade_verts, n_grade_verts:].size > n_grade_verts  # R to L
    # more complicated reversal mapping
    vertices_use = [stc.vertices[0], np.arange(10242)]
    n_src_verts = len(vertices_use[1])
    assert vertices_use[0].shape == (n_grade_verts,)
    assert vertices_use[1].shape == (n_src_verts,)
    # ensure it's sufficiently diffirent to manifest round-trip errors
    assert np.isin(vertices_use[1], stc.vertices[1]).mean() < 0.3
    morph = compute_source_morph(
        stc,
        "fsaverage_sym",
        "fsaverage_sym",
        smooth=smooth,
        xhemi=True,
        warn=False,
        spacing=vertices_use,
        subjects_dir=subjects_dir,
    )
    mm = morph.morph_mat
    assert mm.shape == (n_grade_verts + n_src_verts, n_grade_verts * 2)
    assert mm[:n_grade_verts, :n_grade_verts].size == 0
    assert mm[n_grade_verts:, n_grade_verts:].size == 0
    assert mm[:n_grade_verts, n_grade_verts:].size > n_grade_verts
    assert mm[n_grade_verts:, :n_grade_verts].size > n_src_verts
    # morph forward then back
    stc_xhemi = morph.apply(stc)
    morph = compute_source_morph(
        stc_xhemi,
        "fsaverage_sym",
        "fsaverage_sym",
        smooth=smooth,
        xhemi=True,
        warn=False,
        spacing=stc.vertices,
        subjects_dir=subjects_dir,
    )
    stc_return = morph.apply(stc_xhemi)
    for hi in range(2):
        assert_array_equal(stc_return.vertices[hi], stc.vertices[hi])
    correlation = np.corrcoef(stc.data.ravel(), stc_return.data.ravel())[0, 1]
    assert correlation > 0.9  # not great b/c of sparse grade + small smooth


@testing.requires_testing_data
@pytest.mark.parametrize(
    "smooth, lower, upper, n_warn, dtype",
    [
        (None, 0.959, 0.963, 0, float),
        (3, 0.968, 0.971, 2, complex),
        ("nearest", 0.98, 0.99, 0, float),
    ],
)
def test_surface_source_morph_round_trip(smooth, lower, upper, n_warn, dtype):
    """Test round-trip morphing yields similar STCs."""
    kwargs = dict(smooth=smooth, warn=True, subjects_dir=subjects_dir)
    stc = mne.read_source_estimate(fname_smorph)
    if dtype is complex:
        stc.data = 1j * stc.data
        assert_array_equal(stc.data.real, 0.0)
    with _record_warnings() as w:
        morph = compute_source_morph(stc, "sample", "fsaverage", **kwargs)
    w = [ww for ww in w if "vertices not included" in str(ww.message)]
    assert len(w) == n_warn
    assert morph.morph_mat.shape == (20484, len(stc.data))
    stc_fs = morph.apply(stc)
    morph_back = compute_source_morph(
        stc_fs, "fsaverage", "sample", spacing=stc.vertices, **kwargs
    )
    assert morph_back.morph_mat.shape == (len(stc.data), 20484)
    stc_back = morph_back.apply(stc_fs)
    corr = np.corrcoef(stc.data.ravel(), stc_back.data.ravel())[0, 1]
    assert lower <= corr <= upper
    # check the round-trip power
    assert_power_preserved(stc, stc_back)


@testing.requires_testing_data
def test_surface_source_morph_shortcut():
    """Test that our shortcut for smooth=0 works."""
    stc = mne.read_source_estimate(fname_smorph)
    morph_identity = compute_source_morph(
        stc,
        "sample",
        "sample",
        spacing=stc.vertices,
        smooth=0,
        subjects_dir=subjects_dir,
    )
    stc_back = morph_identity.apply(stc)
    assert_allclose(stc_back.data, stc.data, rtol=1e-4)
    abs_sum = morph_identity.morph_mat - speye(len(stc.data), format="csc")
    abs_sum = np.abs(abs_sum.data).sum()
    assert abs_sum < 1e-4


def assert_power_preserved(orig, new, limits=(1.0, 1.05)):
    """Assert that the power is preserved during a round-trip morph."""
    __tracebackhide__ = True
    for kind in ("real", "imag"):
        numer = np.linalg.norm(getattr(orig.data, kind))
        denom = np.linalg.norm(getattr(new.data, kind))
        if numer == denom == 0.0:  # no data of this type
            continue
        power_ratio = numer / denom
        min_, max_ = limits
        assert min_ < power_ratio < max_, f"Power ratio {kind} = {power_ratio}"


@testing.requires_testing_data
def test_surface_vector_source_morph(tmp_path):
    """Test surface and vector source estimate morph."""
    pytest.importorskip("h5io")
    inverse_operator_surf = read_inverse_operator(fname_inv_surf)

    stc_surf = read_source_estimate(fname_smorph, subject="sample")
    stc_surf.crop(0.09, 0.1)  # for faster computation

    stc_vec = _real_vec_stc()

    source_morph_surf = compute_source_morph(
        inverse_operator_surf["src"], subjects_dir=subjects_dir, smooth=1, warn=False
    )  # smooth 1 for speed
    assert source_morph_surf.subject_from == "sample"
    assert source_morph_surf.subject_to == "fsaverage"
    assert source_morph_surf.kind == "surface"
    assert isinstance(source_morph_surf.src_data, dict)
    assert isinstance(source_morph_surf.src_data["vertices_from"], list)
    assert isinstance(source_morph_surf, SourceMorph)
    stc_surf_morphed = source_morph_surf.apply(stc_surf)
    assert isinstance(stc_surf_morphed, SourceEstimate)
    stc_vec_morphed = source_morph_surf.apply(stc_vec)
    with pytest.raises(ValueError, match="Invalid value for the 'output'"):
        source_morph_surf.apply(stc_surf, output="nifti1")

    # check if correct class after morphing
    assert isinstance(stc_surf_morphed, SourceEstimate)
    assert isinstance(stc_vec_morphed, VectorSourceEstimate)

    # check __repr__
    assert "surface" in repr(source_morph_surf)

    # check loading and saving for surf
    source_morph_surf.save(tmp_path / "42.h5")

    source_morph_surf_r = read_source_morph(tmp_path / "42.h5")

    assert all(
        [
            read == saved
            for read, saved in zip(
                sorted(source_morph_surf_r.__dict__), sorted(source_morph_surf.__dict__)
            )
        ]
    )

    # check wrong subject correction
    stc_surf.subject = None
    assert isinstance(source_morph_surf.apply(stc_surf), SourceEstimate)

    # degenerate
    stc_vol = read_source_estimate(fname_vol_w, "sample")
    with pytest.raises(TypeError, match="stc_from must be an instance"):
        source_morph_surf.apply(stc_vol)


@pytest.mark.slowtest
@testing.requires_testing_data
def test_volume_source_morph_basic(tmp_path):
    """Test volume source estimate morph, special cases and exceptions."""
    nib = pytest.importorskip("nibabel")
    pytest.importorskip("h5io")
    pytest.importorskip("dipy")
    inverse_operator_vol = read_inverse_operator(fname_inv_vol)
    stc_vol = read_source_estimate(fname_vol_w, "sample")

    # check for invalid input type
    with pytest.raises(TypeError, match="src must be"):
        compute_source_morph(src=42)

    # check for raising an error if neither
    # inverse_operator_vol['src'][0]['subject_his_id'] nor subject_from is set,
    # but attempting to perform a volume morph
    src = inverse_operator_vol["src"]
    assert src._subject is None  # already None on disk (old!)

    with pytest.raises(ValueError, match="subject_from could not be inferred"):
        with pytest.warns(RuntimeWarning, match="recommend regenerating"):
            compute_source_morph(src=src, subjects_dir=subjects_dir)

    # check infer subject_from from src[0]['subject_his_id']
    src[0]["subject_his_id"] = "sample"

    with pytest.raises(ValueError, match="Inter-hemispheric morphing"):
        compute_source_morph(src=src, subjects_dir=subjects_dir, xhemi=True)

    with pytest.raises(ValueError, match="Only surface.*sparse morph"):
        compute_source_morph(src=src, sparse=True, subjects_dir=subjects_dir)

    # terrible quality but fast
    zooms = 20
    kwargs = dict(zooms=zooms, niter_sdr=(1,), niter_affine=(1,))
    source_morph_vol = compute_source_morph(
        subjects_dir=subjects_dir, src=fname_inv_vol, subject_from="sample", **kwargs
    )
    shape = (13,) * 3  # for the given zooms

    assert source_morph_vol.subject_from == "sample"

    # the brain used in sample data has shape (255, 255, 255)
    assert tuple(source_morph_vol.sdr_morph.domain_shape) == shape

    assert tuple(source_morph_vol.pre_affine.domain_shape) == shape

    # proofs the above
    assert_array_equal(source_morph_vol.zooms, (zooms,) * 3)

    # assure proper src shape
    mri_size = (src[0]["mri_height"], src[0]["mri_depth"], src[0]["mri_width"])
    assert source_morph_vol.src_data["src_shape_full"] == mri_size

    fwd = read_forward_solution(fname_fwd_vol)
    fwd["src"][0]["subject_his_id"] = "sample"  # avoid further warnings
    source_morph_vol = compute_source_morph(
        fwd["src"], "sample", "sample", subjects_dir=subjects_dir, **kwargs
    )

    # check wrong subject_to
    with pytest.raises(OSError, match="cannot read file"):
        compute_source_morph(fwd["src"], "sample", "42", subjects_dir=subjects_dir)

    # two different ways of saving
    source_morph_vol.save(tmp_path / "vol")

    # check loading
    source_morph_vol_r = read_source_morph(tmp_path / "vol-morph.h5")

    # check for invalid file name handling ()
    with pytest.raises(OSError, match="not found"):
        read_source_morph(tmp_path / "42")

    # check morph
    stc_vol_morphed = source_morph_vol.apply(stc_vol)
    # old way, verts do not match
    assert not np.array_equal(stc_vol_morphed.vertices[0], stc_vol.vertices[0])

    # vector
    stc_vol_vec = VolVectorSourceEstimate(
        np.tile(stc_vol.data[:, np.newaxis], (1, 3, 1)), stc_vol.vertices, 0, 1
    )
    stc_vol_vec_morphed = source_morph_vol.apply(stc_vol_vec)
    assert isinstance(stc_vol_vec_morphed, VolVectorSourceEstimate)
    for ii in range(3):
        assert_allclose(stc_vol_vec_morphed.data[:, ii], stc_vol_morphed.data)

    # check output as NIfTI
    assert isinstance(
        source_morph_vol.apply(stc_vol_vec, output="nifti2"), nib.Nifti2Image
    )

    # check for subject_from mismatch
    source_morph_vol_r.subject_from = "42"
    with pytest.raises(ValueError, match="subject_from must match"):
        source_morph_vol_r.apply(stc_vol_morphed)

    # check if nifti is in grid morph space with voxel_size == spacing
    img_morph_res = source_morph_vol.apply(stc_vol, output="nifti1")

    # assure morph spacing
    assert isinstance(img_morph_res, nib.Nifti1Image)
    assert img_morph_res.header.get_zooms()[:3] == (zooms,) * 3

    # assure src shape
    img_mri_res = source_morph_vol.apply(stc_vol, output="nifti1", mri_resolution=True)
    assert isinstance(img_mri_res, nib.Nifti1Image)
    assert img_mri_res.shape == (
        src[0]["mri_height"],
        src[0]["mri_depth"],
        src[0]["mri_width"],
    ) + (img_mri_res.shape[3],)

    # check if nifti is defined resolution with voxel_size == (5., 5., 5.)
    img_any_res = source_morph_vol.apply(
        stc_vol, output="nifti1", mri_resolution=(5.0, 5.0, 5.0)
    )
    assert isinstance(img_any_res, nib.Nifti1Image)
    assert img_any_res.header.get_zooms()[:3] == (5.0, 5.0, 5.0)

    # check if morph outputs correct data
    assert isinstance(stc_vol_morphed, VolSourceEstimate)

    # check if loaded and saved objects contain the same
    assert all(
        [
            read == saved
            for read, saved in zip(
                sorted(source_morph_vol_r.__dict__), sorted(source_morph_vol.__dict__)
            )
        ]
    )

    # check __repr__
    assert "volume" in repr(source_morph_vol)

    # check Nifti2Image
    assert isinstance(
        source_morph_vol.apply(
            stc_vol, mri_resolution=True, mri_space=True, output="nifti2"
        ),
        nib.Nifti2Image,
    )

    # Degenerate conditions
    with pytest.raises(TypeError, match="output must be"):
        source_morph_vol.apply(stc_vol, output=1)
    with pytest.raises(ValueError, match="subject_from does not match"):
        compute_source_morph(src=src, subject_from="42")
    with pytest.raises(ValueError, match="output"):
        source_morph_vol.apply(stc_vol, output="42")
    with pytest.raises(ValueError, match="subject_to cannot be None"):
        compute_source_morph(src, "sample", None, subjects_dir=subjects_dir)
    # Check if not morphed, but voxel size not boolean, raise ValueError.
    # Note that this check requires dipy to not raise the dipy ImportError
    # before checking if the actual voxel size error will raise.
    with pytest.raises(ValueError, match="Cannot infer original voxel size"):
        stc_vol.as_volume(inverse_operator_vol["src"], mri_resolution=4)

    stc_surf = read_source_estimate(fname_stc, "sample")
    with pytest.raises(TypeError, match="stc_from must be an instance"):
        source_morph_vol.apply(stc_surf)

    # src_to
    source_morph_vol = compute_source_morph(
        fwd["src"],
        subject_from="sample",
        src_to=fwd["src"],
        subject_to="sample",
        subjects_dir=subjects_dir,
        **kwargs,
    )
    stc_vol_2 = source_morph_vol.apply(stc_vol)
    # new way, verts match
    assert_array_equal(stc_vol.vertices[0], stc_vol_2.vertices[0])
    stc_vol_bad = VolSourceEstimate(
        stc_vol.data[:-1], [stc_vol.vertices[0][:-1]], stc_vol.tmin, stc_vol.tstep
    )
    match = (
        "vertices do not match between morph \\(4157\\) and stc \\(4156\\).*"
        "\n.*\n.*\n.*Vertices were likely excluded during forward computatio.*"
    )
    with pytest.raises(ValueError, match=match):
        source_morph_vol.apply(stc_vol_bad)

    # nifti outputs and stc equiv
    img_vol = source_morph_vol.apply(stc_vol, output="nifti1")
    img_vol_2 = stc_vol_2.as_volume(src=fwd["src"], mri_resolution=False)
    assert_allclose(img_vol.affine, img_vol_2.affine)
    img_vol = img_vol.get_fdata()
    img_vol_2 = img_vol_2.get_fdata()
    assert img_vol.shape == img_vol_2.shape
    assert_allclose(img_vol, img_vol_2)


@pytest.mark.slowtest
@testing.requires_testing_data
@pytest.mark.parametrize(
    "subject_from, subject_to, lower, upper, dtype, morph_mat",
    [
        ("sample", "fsaverage", 5.9, 6.1, float, False),
        ("fsaverage", "fsaverage", 0.0, 0.1, float, False),
        ("sample", "sample", 0.0, 0.1, complex, False),
        ("sample", "sample", 0.0, 0.1, float, True),  # morph_mat
        ("sample", "fsaverage", 10, 12, float, True),  # morph_mat
    ],
)
def test_volume_source_morph_round_trip(
    tmp_path, subject_from, subject_to, lower, upper, dtype, morph_mat, monkeypatch
):
    """Test volume source estimate morph round-trips well."""
    nib = pytest.importorskip("nibabel")
    pytest.importorskip("h5io")
    pytest.importorskip("dipy")
    from nibabel.processing import resample_from_to

    src = dict()
    if morph_mat:
        # ~1.5 minutes with pos=7. (4157 morphs!) for sample, so only test
        # morph_mat computation mode with a few labels
        label_names = sorted(get_volume_labels_from_aseg(fname_aseg))[1:2]
        if "sample" in (subject_from, subject_to):
            src["sample"] = setup_volume_source_space(
                "sample",
                subjects_dir=subjects_dir,
                volume_label=label_names,
                mri=fname_aseg,
            )
            assert sum(s["nuse"] for s in src["sample"]) == 12
        if "fsaverage" in (subject_from, subject_to):
            src["fsaverage"] = setup_volume_source_space(
                "fsaverage",
                subjects_dir=subjects_dir,
                volume_label=label_names[:3],
                mri=fname_aseg_fs,
            )
            assert sum(s["nuse"] for s in src["fsaverage"]) == 16
    else:
        assert not morph_mat
        if "sample" in (subject_from, subject_to):
            src["sample"] = mne.read_source_spaces(fname_vol)
            src["sample"][0]["subject_his_id"] = "sample"
            assert src["sample"][0]["nuse"] == 4157
        if "fsaverage" in (subject_from, subject_to):
            # Created to save space with:
            #
            # bem = op.join(op.dirname(mne.__file__), 'data', 'fsaverage',
            #               'fsaverage-inner_skull-bem.fif')
            # src_fsaverage = mne.setup_volume_source_space(
            #     'fsaverage', pos=7., bem=bem, mindist=0,
            #     subjects_dir=subjects_dir, add_interpolator=False)
            # mne.write_source_spaces(fname_fs_vol, src_fsaverage,
            #                         overwrite=True)
            #
            # For speed we do it without the interpolator because it's huge.
            src["fsaverage"] = mne.read_source_spaces(fname_fs_vol)
            src["fsaverage"][0].update(
                vol_dims=np.array([23, 29, 25]), seg_name="brain"
            )
            _add_interpolator(src["fsaverage"])
            assert src["fsaverage"][0]["nuse"] == 6379
    src_to, src_from = src[subject_to], src[subject_from]
    del src
    # No SDR just for speed once everything works
    kwargs = dict(
        niter_sdr=(), niter_affine=(1,), subjects_dir=subjects_dir, verbose=True
    )
    morph_from_to = compute_source_morph(
        src=src_from, src_to=src_to, subject_to=subject_to, **kwargs
    )
    morph_to_from = compute_source_morph(
        src=src_to, src_to=src_from, subject_to=subject_from, **kwargs
    )
    nuse = sum(s["nuse"] for s in src_from)
    assert nuse > 10
    use = np.linspace(0, nuse - 1, 10).round().astype(int)
    data = np.eye(nuse)[:, use]
    if dtype is complex:
        data = data * 1j
    vertices = [s["vertno"] for s in src_from]
    stc_from = VolSourceEstimate(data, vertices, 0, 1)
    with catch_logging() as log:
        stc_from_rt = morph_to_from.apply(
            morph_from_to.apply(stc_from, verbose="debug")
        )
    log = log.getvalue()
    assert "individual volume morph" in log
    maxs = np.argmax(stc_from_rt.data, axis=0)
    src_rr = np.concatenate([s["rr"][s["vertno"]] for s in src_from])
    dists = 1000 * np.linalg.norm(src_rr[use] - src_rr[maxs], axis=1)
    mu = np.mean(dists)
    # fsaverage=5.99; 7.97 without additional src_ras_t fix
    # fsaverage=7.97; 25.4 without src_ras_t fix
    assert lower <= mu < upper, f"round-trip distance {mu}"
    # check that pre_affine is close to identity when subject_to==subject_from
    if subject_to == subject_from:
        for morph in (morph_to_from, morph_from_to):
            assert_allclose(morph.pre_affine.affine, np.eye(4), atol=1e-2)
    # check that power is more or less preserved (labelizing messes with this)
    if morph_mat:
        if subject_to == "fsaverage":
            limits = (18, 18.5)
        else:
            limits = (7, 7.5)
    else:
        limits = (1, 1.2)
    stc_from_unit = stc_from.copy().crop(0, 0)
    stc_from_unit._data.fill(1.0)
    stc_from_unit_rt = morph_to_from.apply(morph_from_to.apply(stc_from_unit))
    assert_power_preserved(stc_from_unit, stc_from_unit_rt, limits=limits)
    if morph_mat:
        fname = tmp_path / "temp-morph.h5"
        morph_to_from.save(fname)
        morph_to_from = read_source_morph(fname)
        assert morph_to_from.vol_morph_mat is None
        morph_to_from.compute_vol_morph_mat(verbose=True)
        morph_to_from.save(fname, overwrite=True)
        morph_to_from = read_source_morph(fname)
        assert isinstance(morph_to_from.vol_morph_mat, csr_array), "csr"
        # equivalence (plus automatic calling)
        assert morph_from_to.vol_morph_mat is None
        monkeypatch.setattr(mne.morph, "_VOL_MAT_CHECK_RATIO", 0.0)
        with catch_logging() as log:
            with pytest.warns(RuntimeWarning, match=r"calling morph\.compute"):
                stc_from_rt_lin = morph_to_from.apply(
                    morph_from_to.apply(stc_from, verbose="debug")
                )
        assert isinstance(morph_from_to.vol_morph_mat, csr_array), "csr"
        log = log.getvalue()
        assert "sparse volume morph matrix" in log
        assert_allclose(stc_from_rt.data, stc_from_rt_lin.data)
        del stc_from_rt_lin
        stc_from_unit_rt_lin = morph_to_from.apply(morph_from_to.apply(stc_from_unit))
        assert_allclose(stc_from_unit_rt.data, stc_from_unit_rt_lin.data)
        del stc_from_unit_rt_lin
    del stc_from, stc_from_rt
    # before and after morph, check the proportion of vertices
    # that are inside and outside the brainmask.mgz
    brain = nib.load(subjects_dir / subject_from / "mri" / "brain.mgz")
    mask = _get_img_fdata(brain) > 0
    if subject_from == subject_to == "sample":
        for stc in [stc_from_unit, stc_from_unit_rt]:
            img = stc.as_volume(src_from, mri_resolution=True)
            img = nib.Nifti1Image(  # abs to convert complex
                np.abs(_get_img_fdata(img)[:, :, :, 0]), img.affine
            )
            img = _get_img_fdata(resample_from_to(img, brain, order=1))
            assert img.shape == mask.shape
            in_ = img[mask].astype(bool).mean()
            out = img[~mask].astype(bool).mean()
            if morph_mat:
                out_max = 0.001
                in_min, in_max = 0.005, 0.007
            else:
                out_max = 0.02
                in_min, in_max = 0.97, 0.98
            assert out < out_max, f"proportion out of volume {out}"
            assert in_min < in_ < in_max, f"proportion inside volume {in_}"


@pytest.mark.slowtest
@testing.requires_testing_data
def test_morph_stc_dense():
    """Test morphing stc."""
    subject_from = "sample"
    subject_to = "fsaverage"
    stc_from = read_source_estimate(fname_smorph, subject="sample")
    stc_to = read_source_estimate(fname_fmorph)
    # make sure we can specify grade
    stc_from.crop(0.09, 0.1)  # for faster computation
    stc_to.crop(0.09, 0.1)  # for faster computation
    assert_array_equal(
        stc_to.time_as_index([0.09, 0.1], use_rounding=True), [0, len(stc_to.times) - 1]
    )

    # After dep change this to:
    morph = compute_source_morph(
        subject_to=subject_to,
        spacing=3,
        smooth=12,
        src=stc_from,
        subjects_dir=subjects_dir,
        precompute=True,
    )
    assert morph.vol_morph_mat is None  # a no-op for surface
    stc_to1 = morph.apply(stc_from)
    assert_allclose(stc_to.data, stc_to1.data, atol=1e-5)

    mean_from = stc_from.data.mean(axis=0)
    mean_to = stc_to1.data.mean(axis=0)
    assert np.corrcoef(mean_to, mean_from).min() > 0.999

    vertices_to = grade_to_vertices(subject_to, grade=3, subjects_dir=subjects_dir)

    # make sure we can fill by morphing
    with pytest.warns(RuntimeWarning, match="consider increasing"):
        morph = compute_source_morph(
            stc_from,
            subject_from,
            subject_to,
            spacing=None,
            smooth=1,
            subjects_dir=subjects_dir,
        )
    stc_to5 = morph.apply(stc_from)
    assert stc_to5.data.shape[0] == 163842 + 163842

    # Morph vector data
    stc_vec = _real_vec_stc()
    stc_vec_to1 = compute_source_morph(
        stc_vec,
        subject_from,
        subject_to,
        subjects_dir=subjects_dir,
        spacing=vertices_to,
        smooth=1,
        warn=False,
    ).apply(stc_vec)
    assert stc_vec_to1.subject == subject_to
    assert stc_vec_to1.tmin == stc_vec.tmin
    assert stc_vec_to1.tstep == stc_vec.tstep
    assert len(stc_vec_to1.lh_vertno) == 642
    assert len(stc_vec_to1.rh_vertno) == 642

    # Degenerate conditions

    # Morphing to a density that is too high should raise an informative error
    # (here we need to push to grade=6, but for some subjects even grade=5
    # will break)
    with pytest.raises(ValueError, match="Cannot use icosahedral grade 6 "):
        compute_source_morph(
            stc_to1,
            subject_from=subject_to,
            subject_to=subject_from,
            spacing=6,
            subjects_dir=subjects_dir,
        )
    del stc_to1

    with pytest.raises(ValueError, match="smooth.* has to be at least 0"):
        compute_source_morph(
            stc_from,
            subject_from,
            subject_to,
            spacing=5,
            smooth=-1,
            subjects_dir=subjects_dir,
        )

    # subject from mismatch
    with pytest.raises(ValueError, match="subject_from does not match"):
        compute_source_morph(stc_from, subject_from="foo", subjects_dir=subjects_dir)

    # only one set of vertices
    with pytest.raises(ValueError, match="grade.*list must have two elements"):
        compute_source_morph(
            stc_from,
            subject_from=subject_from,
            spacing=[vertices_to[0]],
            subjects_dir=subjects_dir,
        )


@testing.requires_testing_data
def test_morph_stc_sparse():
    """Test morphing stc with sparse=True."""
    subject_from = "sample"
    subject_to = "fsaverage"
    # Morph sparse data
    # Make a sparse stc
    stc_from = read_source_estimate(fname_smorph, subject="sample")
    stc_from.vertices[0] = stc_from.vertices[0][[100, 500]]
    stc_from.vertices[1] = stc_from.vertices[1][[200]]
    stc_from._data = stc_from._data[:3]

    stc_to_sparse = compute_source_morph(
        stc_from,
        subject_from=subject_from,
        subject_to=subject_to,
        spacing=None,
        sparse=True,
        subjects_dir=subjects_dir,
    ).apply(stc_from)

    assert_allclose(
        np.sort(stc_from.data.sum(axis=1)), np.sort(stc_to_sparse.data.sum(axis=1))
    )
    assert len(stc_from.rh_vertno) == len(stc_to_sparse.rh_vertno)
    assert len(stc_from.lh_vertno) == len(stc_to_sparse.lh_vertno)
    assert stc_to_sparse.subject == subject_to
    assert stc_from.tmin == stc_from.tmin
    assert stc_from.tstep == stc_from.tstep

    stc_from.vertices[0] = np.array([], dtype=np.int64)
    stc_from._data = stc_from._data[:1]

    stc_to_sparse = compute_source_morph(
        stc_from,
        subject_from,
        subject_to,
        spacing=None,
        sparse=True,
        subjects_dir=subjects_dir,
    ).apply(stc_from)

    assert_allclose(
        np.sort(stc_from.data.sum(axis=1)), np.sort(stc_to_sparse.data.sum(axis=1))
    )
    assert len(stc_from.rh_vertno) == len(stc_to_sparse.rh_vertno)
    assert len(stc_from.lh_vertno) == len(stc_to_sparse.lh_vertno)
    assert stc_to_sparse.subject == subject_to
    assert stc_from.tmin == stc_from.tmin
    assert stc_from.tstep == stc_from.tstep

    # Degenerate cases
    with pytest.raises(ValueError, match="spacing must be set to None"):
        compute_source_morph(
            stc_from,
            subject_from=subject_from,
            subject_to=subject_to,
            spacing=5,
            sparse=True,
            subjects_dir=subjects_dir,
        )
    with pytest.raises(ValueError, match="xhemi=True can only be used with"):
        compute_source_morph(
            stc_from,
            subject_from=subject_from,
            subject_to=subject_to,
            spacing=None,
            sparse=True,
            xhemi=True,
            subjects_dir=subjects_dir,
        )


@testing.requires_testing_data
@pytest.mark.parametrize(
    "sl, n_real, n_mri, n_orig",
    [
        # First and last should add up, middle can have overlap should be <= sum
        (slice(0, 1), 37, 138, 8),
        (slice(1, 2), 51, 204, 12),
        (slice(0, 2), 88, 324, 20),
    ],
)
def test_volume_labels_morph(tmp_path, sl, n_real, n_mri, n_orig):
    """Test generating a source space from volume label."""
    nib = pytest.importorskip("nibabel")
    n_use = (sl.stop - sl.start) // (sl.step or 1)
    # see gh-5224
    evoked = mne.read_evokeds(fname_evoked)[0].crop(0, 0)
    evoked.pick(evoked.ch_names[:306:8])
    evoked.info.normalize_proj()
    n_ch = len(evoked.ch_names)
    lut, _ = read_freesurfer_lut()
    label_names = sorted(get_volume_labels_from_aseg(fname_aseg))
    use_label_names = label_names[sl]
    src = setup_volume_source_space(
        "sample",
        subjects_dir=subjects_dir,
        volume_label=use_label_names,
        mri=fname_aseg,
    )
    assert len(src) == n_use
    assert src.kind == "volume"
    n_src = sum(s["nuse"] for s in src)
    sphere = make_sphere_model("auto", "auto", evoked.info)
    fwd = make_forward_solution(evoked.info, fname_trans, src, sphere)
    assert fwd["sol"]["data"].shape == (n_ch, n_src * 3)
    inv = make_inverse_operator(
        evoked.info, fwd, make_ad_hoc_cov(evoked.info), loose=1.0
    )
    stc = apply_inverse(evoked, inv)
    assert stc.data.shape == (n_src, 1)
    img = stc.as_volume(src, mri_resolution=True)
    assert img.shape == (86, 86, 86, 1)
    n_on = np.array(img.dataobj).astype(bool).sum()
    aseg_img = _get_img_fdata(nib.load(fname_aseg))
    n_got_real = np.isin(
        aseg_img.ravel(), [lut[name] for name in use_label_names]
    ).sum()
    assert n_got_real == n_real
    # - This was 291 on `main` before gh-5590
    # - Refactoring transforms it became 279 with a < 1e-8 change in vox_mri_t
    # - Dropped to 123 once nearest-voxel was used in gh-7653
    # - Jumped back up to 330 with morphing fixes actually correctly
    #   interpolating across all volumes
    assert aseg_img.shape == img.shape[:3]
    assert n_on == n_mri
    for ii in range(2):
        # should work with (ii=0) or without (ii=1) the interpolator
        if ii:
            src[0]["interpolator"] = None
        img = stc.as_volume(src, mri_resolution=False)
        n_on = np.array(img.dataobj).astype(bool).sum()
        # was 20 on `main` before gh-5590
        # then 44 before gh-7653, which took it back to 20
        assert n_on == n_orig
    # without the interpolator, this should fail
    assert src[0]["interpolator"] is None
    with pytest.raises(RuntimeError, match=r".*src\[0\], .* mri_resolution"):
        stc.as_volume(src, mri_resolution=True)


@pytest.fixture(scope="session", params=[testing._pytest_param()])
def _mixed_morph_srcs():
    pytest.importorskip("nibabel")
    pytest.importorskip("dipy")
    # create a mixed source space
    labels_vol = ["Left-Cerebellum-Cortex", "Right-Cerebellum-Cortex"]
    src = mne.setup_source_space(
        "sample", spacing="oct3", add_dist=False, subjects_dir=subjects_dir
    )
    src += mne.setup_volume_source_space(
        "sample",
        mri=fname_aseg,
        pos=10.0,
        volume_label=labels_vol,
        subjects_dir=subjects_dir,
        add_interpolator=True,
        verbose=True,
    )
    # create the destination space
    src_fs = mne.read_source_spaces(
        subjects_dir / "fsaverage" / "bem" / "fsaverage-ico-5-src.fif"
    )
    src_fs += mne.setup_volume_source_space(
        "fsaverage",
        pos=7.0,
        volume_label=labels_vol,
        subjects_dir=subjects_dir,
        add_interpolator=False,
        verbose=True,
    )
    del labels_vol

    with pytest.raises(ValueError, match="src_to must be provided .* mixed"):
        mne.compute_source_morph(
            src=src,
            subject_from="sample",
            subject_to="fsaverage",
            subjects_dir=subjects_dir,
        )

    with pytest.warns(RuntimeWarning, match="not included in smoothing"):
        morph = mne.compute_source_morph(
            src=src,
            subject_from="sample",
            subject_to="fsaverage",
            subjects_dir=subjects_dir,
            niter_affine=[1, 0, 0],
            niter_sdr=[1, 0, 0],
            src_to=src_fs,
            smooth=5,
            verbose=True,
        )
    return morph, src, src_fs


@pytest.mark.slowtest
@pytest.mark.parametrize("vector", (False, True))
def test_mixed_source_morph(_mixed_morph_srcs, vector):
    """Test mixed source space morphing."""
    nib = pytest.importorskip("nibabel")
    pytest.importorskip("dipy")
    morph, src, src_fs = _mixed_morph_srcs
    # Test some basic properties in the subject's own space
    lut, _ = read_freesurfer_lut()
    ids = [lut[s["seg_name"]] for s in src[2:]]
    del lut
    vertices = [s["vertno"] for s in src]
    n_vertices = sum(len(v) for v in vertices)
    data = np.zeros((n_vertices, 3, 1))
    data[:, 1] = 1.0
    klass = mne.MixedVectorSourceEstimate
    if not vector:
        data = data[:, 1]
        klass = klass._scalar_class
    stc = klass(data, vertices, 0, 1, "sample")
    vol_info = _get_mri_info_data(fname_aseg, data=True)
    rrs = np.concatenate([src[2]["rr"][sp["vertno"]] for sp in src[2:]])
    n_want = np.isin(_get_atlas_values(vol_info, rrs), ids).sum()
    img = _get_img_fdata(stc.volume().as_volume(src, mri_resolution=False))
    assert img.astype(bool).sum() == n_want
    img_res = nib.load(fname_aseg)
    n_want = np.isin(_get_img_fdata(img_res), ids).sum()
    img = _get_img_fdata(stc.volume().as_volume(src, mri_resolution=True))
    assert img.astype(bool).sum() > n_want  # way more get interpolated into

    with pytest.raises(TypeError, match="stc_from must be an instance"):
        morph.apply(1.0)

    # Now actually morph
    stc_fs = morph.apply(stc)
    vol_info = _get_mri_info_data(fname_aseg_fs, data=True)
    rrs = np.concatenate([src_fs[2]["rr"][sp["vertno"]] for sp in src_fs[2:]])
    n_want = np.isin(_get_atlas_values(vol_info, rrs), ids).sum()
    with pytest.raises(ValueError, match=r"stc\.subject does not match src s"):
        stc_fs.volume().as_volume(src, mri_resolution=False)
    img = _get_img_fdata(stc_fs.volume().as_volume(src_fs, mri_resolution=False))
    assert img.astype(bool).sum() == n_want  # correct number of voxels

    # Morph separate parts and compare to morphing the entire one
    stc_fs_surf = morph.apply(stc.surface())
    stc_fs_vol = morph.apply(stc.volume())
    stc_fs_2 = stc_fs.__class__(
        np.concatenate([stc_fs_surf.data, stc_fs_vol.data]),
        stc_fs_surf.vertices + stc_fs_vol.vertices,
        stc_fs.tmin,
        stc_fs.tstep,
        stc_fs.subject,
    )
    assert_allclose(stc_fs.data, stc_fs_2.data)


def _rand_affine(rng):
    quat = rng.randn(3)
    quat /= 5 * np.linalg.norm(quat)
    affine = np.eye(4)
    affine[:3, 3] = rng.randn(3) / 5.0
    affine[:3, :3] = quat_to_rot(quat)
    return affine


_shapes = (
    (10, 10, 10),
    (20, 5, 10),
    (5, 10, 20),
)
_affines = (
    [[2, 0, 0, 1], [0, 0, 1, -1], [0, -1, 0, 2], [0, 0, 0, 1]],
    np.eye(4),
    np.eye(4)[[0, 2, 1, 3]],
    "rand",
)


@pytest.mark.parametrize("from_shape", _shapes)
@pytest.mark.parametrize("from_affine", _affines)
@pytest.mark.parametrize("to_shape", _shapes)
@pytest.mark.parametrize("to_affine", _affines)
@pytest.mark.parametrize("order", [0, 1])
@pytest.mark.parametrize("seed", [0, 1])
def test_resample_equiv(from_shape, from_affine, to_shape, to_affine, order, seed):
    """Test resampling equivalences."""
    pytest.importorskip("nibabel")
    pytest.importorskip("dipy")
    rng = np.random.RandomState(seed)
    from_data = rng.randn(*from_shape)
    is_rand = False
    if isinstance(to_affine, str):
        assert to_affine == "rand"
        to_affine = _rand_affine(rng)
        is_rand = True
    if isinstance(from_affine, str):
        assert from_affine == "rand"
        from_affine = _rand_affine(rng)
        is_rand = True
    to_affine = np.array(to_affine, float)
    assert to_affine.shape == (4, 4)
    from_affine = np.array(from_affine, float)
    assert from_affine.shape == (4, 4)
    #
    # 1. nibabel.processing.resample_from_to
    #
    # for a 1mm iso / 256 -> 5mm / 51 one sample takes ~486 ms
    from nibabel.processing import resample_from_to
    from nibabel.spatialimages import SpatialImage

    start = np.linalg.norm(from_data)
    got_nibabel = resample_from_to(
        SpatialImage(from_data, from_affine), (to_shape, to_affine), order=order
    ).get_fdata()
    end = np.linalg.norm(got_nibabel)
    assert end > 0.05 * start  # not too much power lost
    #
    # 2. dipy.align.imaffine
    #
    # ~366 ms
    import dipy.align.imaffine

    interp = "linear" if order == 1 else "nearest"
    got_dipy = dipy.align.imaffine.AffineMap(
        None,
        domain_grid_shape=to_shape,
        domain_grid2world=to_affine,
        codomain_grid_shape=from_shape,
        codomain_grid2world=from_affine,
    ).transform(from_data, interpolation=interp, resample_only=True)
    # XXX possibly some error in dipy or nibabel (/SciPy), or some boundary
    # condition?
    nib_different = (is_rand and order == 1) or (
        from_affine[0, 0] == 2.0 and not np.allclose(from_affine, to_affine)
    )
    nib_different = nib_different and not (
        is_rand and from_affine[0, 0] == 2 and order == 0
    )
    if nib_different:
        assert not np.allclose(got_dipy, got_nibabel), "nibabel fixed"
    else:
        assert_allclose(got_dipy, got_nibabel, err_msg="dipy<->nibabel")
    #
    # 3. mne.source_space._grid_interp
    #
    # ~339 ms
    trans = np.linalg.inv(from_affine) @ to_affine  # to -> from
    interp = _grid_interp(from_shape, to_shape, trans, order=order)
    got_mne = np.asarray(interp @ from_data.ravel(order="F")).reshape(
        to_shape, order="F"
    )
    if order == 1:
        assert_allclose(got_mne, got_dipy, err_msg="MNE<->dipy")
    else:
        perc = 100 * np.isclose(got_mne, got_dipy).mean()
        assert 83 < perc <= 100