%global srcname HD-BET
%global import_name HD_BET

Name:           python-hd-bet
Version:        2.0.1
Release:        1%{?dist}
Summary:        Brain extraction for multi-sequence MRI with nnU-Net

License:        Apache-2.0
URL:            https://github.com/MIC-DKFZ/HD-BET
# The GitHub tag rather than the PyPI sdist: it carries the readme, which the
# sdist leaves out although pyproject.toml names it.
Source0:        %{url}/archive/v%{version}/%{srcname}-%{version}.tar.gz

BuildArch:      noarch
BuildRequires:  python3-devel
BuildRequires:  pyproject-rpm-macros
# %%check runs apply_bet on a synthetic volume and the CLI's --help
BuildRequires:  python3dist(simpleitk)

%global _description %{expand:
HD-BET removes the skull and other non-brain tissue from MRI of the head. It
was trained on T1, contrast-enhanced T1, T2 and FLAIR scans of brain-tumour
patients from many scanners and centres, and handles pathology and post-
operative anatomy that classical brain extraction tools often fail on. Since
version 2.0 it is a thin wrapper around nnU-Net v2 inference.

The trained network (114 MB, CC BY-NC 4.0, Zenodo record 14445620) is not part
of this package: the hd-bet command downloads it on first use into
~/hd-bet_params. Pass "-device cpu" on machines without a CUDA GPU.}

%description %_description

%package -n python3-hd-bet
Summary:        %{summary}
%description -n python3-hd-bet %_description

%prep
%autosetup -n %{srcname}-%{version}
# pyproject.toml names README.md; the file in the tag is readme.md.
sed -i 's/^readme = "README.md"/readme = "readme.md"/' pyproject.toml
grep -q '^readme = "readme.md"' pyproject.toml

%generate_buildrequires
%pyproject_buildrequires

%build
%pyproject_wheel

%install
%pyproject_install
%pyproject_save_files %{import_name}

%check
%pyproject_check_import

# The CLI parses its arguments before it touches the network.
%{buildroot}%{_bindir}/hd-bet --help | grep -q -- '--disable_tta'

# apply_bet is the post-processing step that turns the network's mask into the
# brain-extracted image. Check it keeps voxels, zeroes the rest and carries the
# geometry over, on a synthetic volume (the network itself needs the weights).
cd %{_builddir}
%{py3_test_envvars} %{python3} - <<'EOF'
import numpy as np, SimpleITK as sitk
from HD_BET.hd_bet_prediction import apply_bet
img = np.arange(10*12*14, dtype=np.float32).reshape(10, 12, 14) + 1
mask = np.zeros_like(img, dtype=np.uint8); mask[2:8, 3:9, 4:10] = 1
i = sitk.GetImageFromArray(img); i.SetSpacing((0.9, 1.1, 1.3)); i.SetOrigin((-5, 7, 3))
m = sitk.GetImageFromArray(mask); m.CopyInformation(i)
sitk.WriteImage(i, "t1.nii.gz"); sitk.WriteImage(m, "t1_bet.nii.gz")
apply_bet("t1.nii.gz", "t1_bet.nii.gz", "out.nii.gz")
o = sitk.ReadImage("out.nii.gz"); a = sitk.GetArrayFromImage(o)
assert np.array_equal(a[mask == 1], img[mask == 1])
assert not a[mask == 0].any()
assert np.allclose(o.GetSpacing(), (0.9, 1.1, 1.3)) and np.allclose(o.GetOrigin(), (-5, 7, 3))
print("apply_bet ok:", int(mask.sum()), "voxels kept")
EOF

%files -n python3-hd-bet -f %{pyproject_files}
%license LICENSE
%doc readme.md
%{_bindir}/hd-bet

%changelog
* Fri Oct 16 2026 Morgan Hough <morgan.hough@gmail.com> - 2.0.1-1
- Initial package. The CC BY-NC 4.0 weights stay a first-use download.