segauge 0.1.0__tar.gz

segauge-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,28 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.10", "3.11", "3.12", "3.13"]
+    steps:
+      - uses: actions/checkout@v4
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Sync dependencies
+        run: uv sync --extra dicom
+      - name: Lint
+        run: |
+          uv run ruff check .
+          uv run ruff format --check .
+      - name: Test
+        run: uv run pytest -q

segauge-0.1.0/.github/workflows/docs.yml

@@ -0,0 +1,22 @@
+name: docs
+
+on:
+  push:
+    branches: [main]
+
+permissions:
+  contents: write
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          python-version: "3.12"
+      - name: Build and deploy docs
+        run: |
+          uv sync --group docs
+          uv run mkdocs gh-deploy --force

segauge-0.1.0/.gitignore

@@ -0,0 +1,29 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+.eggs/
+build/
+dist/
+.venv/
+venv/
+
+# Tooling
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+.coverage
+htmlcov/
+coverage.xml
+
+# Editors / OS
+.vscode/
+.idea/
+.DS_Store
+
+# Local artifacts
+*.html
+!src/segauge/report/templates/*.html
+scratch/
+tasks/
+site/

segauge-0.1.0/CITATION.cff

@@ -0,0 +1,18 @@
+cff-version: 1.2.0
+message: "If you use segauge in your research, please cite it."
+title: "segauge: geometrically-correct, DICOM-native evaluation metrics for medical image segmentation"
+authors:
+  - family-names: Haisma
+    given-names: Ruben
+type: software
+license: Apache-2.0
+repository-code: "https://github.com/RubenHaisma/segauge"
+url: "https://rubenhaisma.github.io/segauge/"
+keywords:
+  - medical image segmentation
+  - evaluation metrics
+  - Hausdorff distance
+  - HD95
+  - normalized surface dice
+  - DICOM
+  - Python

segauge-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or Derivative
+          Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Ruben Haisma
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

segauge-0.1.0/PKG-INFO

@@ -0,0 +1,117 @@
+Metadata-Version: 2.4
+Name: segauge
+Version: 0.1.0
+Summary: The honest, geometrically-correct, DICOM-native evaluation harness for medical image segmentation.
+Project-URL: Homepage, https://github.com/RubenHaisma/segauge
+Project-URL: Repository, https://github.com/RubenHaisma/segauge
+Project-URL: Issues, https://github.com/RubenHaisma/segauge/issues
+Author: Ruben Haisma
+License: Apache-2.0
+License-File: LICENSE
+Keywords: assd,computer-vision,confidence-interval,dice,dicom,dicom-seg,evaluation,hausdorff,hd95,image-segmentation,iou,jaccard,medical-image-analysis,medical-imaging,metrics,monai,nifti,nnunet,nsd,radiology,rtstruct,segmentation,surface-dice,surface-distance,totalsegmentator
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Healthcare Industry
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Image Processing
+Classifier: Topic :: Scientific/Engineering :: Medical Science Apps.
+Requires-Python: >=3.10
+Requires-Dist: jinja2>=3.1
+Requires-Dist: nibabel>=5.0
+Requires-Dist: numpy>=1.24
+Requires-Dist: rtree>=1.0
+Requires-Dist: scikit-image>=0.21
+Requires-Dist: scipy>=1.10
+Requires-Dist: trimesh>=4.0
+Provides-Extra: dicom
+Requires-Dist: highdicom>=0.22; extra == 'dicom'
+Requires-Dist: pydicom>=2.4; extra == 'dicom'
+Requires-Dist: rt-utils>=1.2; extra == 'dicom'
+Description-Content-Type: text/markdown
+
+# segauge
+
+**DICOM-native evaluation for medical image segmentation: surface-mesh distance metrics, per-lesion detection, fairness slices, and a confidence interval on every number.**
+
+[![PyPI](https://img.shields.io/pypi/v/segauge.svg)](https://pypi.org/project/segauge/)
+[![Python](https://img.shields.io/pypi/pyversions/segauge.svg)](https://pypi.org/project/segauge/)
+[![License](https://img.shields.io/badge/license-Apache--2.0-blue.svg)](LICENSE)
+
+segauge computes Dice, IoU, Hausdorff distance (HD), **HD95**, average symmetric surface distance (**ASSD**), **Normalized Surface Dice (NSD)**, and **per-lesion detection F1** for 3D medical image segmentation, and puts a **bootstrap confidence interval** on every number. It reads **NIfTI, DICOM-SEG, RTSTRUCT, and NumPy** directly, so you can evaluate the output of nnU-Net, MONAI, or TotalSegmentator without a lossy conversion.
+
+It exists because the metrics you report should be the metrics that are true, and today they often aren't.
+
+```bash
+pip install segauge
+```
+
+```python
+import segauge as sg
+
+result = sg.evaluate([
+    sg.Case("patient_001", pred="pred.nii.gz", gt="gt.nii.gz", metadata={"scanner": "siemens"}),
+    sg.Case("patient_002", pred="pred2.dcm",   gt="gt2.dcm",   metadata={"scanner": "ge"}),
+])
+
+print(result.summary())          # every metric with a 95% CI
+print(result.by_subgroup("scanner"))   # where does the model quietly fail?
+result.to_html("report.html")    # one self-contained report
+```
+
+Or from the command line:
+
+```bash
+segauge eval --pred preds/ --gt labels/ --metadata cases.csv --report report.html
+```
+
+## Why not just use MONAI or Metrics Reloaded?
+
+segauge bundles four things no incumbent provides together: a **confidence interval** on every metric, **per-lesion detection**, **subgroup/fairness slicing**, and **native DICOM-SEG/RTSTRUCT** input.
+
+It also computes distance metrics on a surface mesh rather than the voxel grid: it extracts the object surface with marching cubes **at true voxel spacing** and integrates over the surface, following the [MeshMetrics](https://arxiv.org/abs/2509.05670) method (Podobnik & Vrtovec, 2025). In our own benchmark (`benchmarks/mesh_vs_grid.py`) this measurably reduces **HD95** error under anisotropic, thick-slice spacing, where grid rasterization hurts most; for mean distance (ASSD) on smooth shapes the two methods are comparable. A formal validation suite against the MeshMetrics reference is in progress; we report what the benchmark shows, not more.
+
+| | segauge | MONAI | Metrics Reloaded | seg-metrics | DeepMind surface-distance |
+|---|:---:|:---:|:---:|:---:|:---:|
+| Surface-mesh distances (vs voxel grid) | ✅ | ❌ | ❌ | ❌ | ❌ |
+| Confidence intervals | ✅ | ❌ | ❌ | ❌ | ❌ |
+| Per-lesion detection F1 | ✅ | partial | ✅ | ❌ | ❌ |
+| Subgroup / fairness slicing | ✅ | ❌ | ❌ | ❌ | ❌ |
+| DICOM-SEG / RTSTRUCT native | ✅ | partial | ❌ | ❌ | ❌ |
+| `pip install` | ✅ | ✅ | ❌ | ✅ | ✅ |
+
+## Metrics
+
+- **Overlap:** Dice (DSC), IoU (Jaccard) — exact, integer-counted.
+- **Surface distance:** Hausdorff (HD), HD95, ASSD, MASD, Normalized Surface Dice (NSD) — mesh-based, spacing-aware, area-weighted.
+- **Per-lesion detection:** precision, recall, F1 via connected-component matching, so you can answer "did it find the tumor?" not just "how much voxel overlap?"
+- Every aggregate carries a deterministic **bootstrap confidence interval**.
+
+## Inputs
+
+NIfTI (`.nii`, `.nii.gz`), DICOM-SEG, RTSTRUCT, NumPy arrays, and `.npy`. Voxel spacing is read from the file header and used for the distance metrics, so a segmentation from a clinical pipeline is evaluated as-is.
+
+## FAQ
+
+**How do I compute HD95 correctly in Python?** `segauge.surface_metrics(pred, gt, spacing)` returns HD, HD95, ASSD, MASD, and NSD computed on the surface mesh, not the voxel grid.
+
+**Can it evaluate DICOM-SEG / RTSTRUCT directly?** Yes. `pip install segauge[dicom]` and pass a `.dcm` SEG, or use `segauge.load_rtstruct(series_dir, rtstruct, roi_name)`.
+
+**Does it work with nnU-Net / TotalSegmentator / MONAI outputs?** Yes. Point it at the NIfTI (or DICOM) they produce.
+
+**Why a confidence interval?** A Dice of 0.85 on 12 cases is not the same claim as 0.85 on 1200. segauge makes the difference visible.
+
+**2D images?** Overlap and detection metrics work in any dimension; surface-distance metrics are 3D in v0.1 (2D is planned for v0.2).
+
+## Status
+
+Pre-release (`0.1.0.dev`), built in the open. segauge is an evaluation tool for developers and researchers. It is **not a medical device** and produces no diagnosis.
+
+## References
+
+- Maier-Hein, Reinke et al. *Metrics Reloaded.* Nature Methods (2024). https://www.nature.com/articles/s41592-023-02151-z
+- Podobnik & Vrtovec. *MeshMetrics.* arXiv:2509.05670 (2025). https://arxiv.org/abs/2509.05670
+
+## License
+
+Apache-2.0.

segauge-0.1.0/README.md

@@ -0,0 +1,85 @@
+# segauge
+
+**DICOM-native evaluation for medical image segmentation: surface-mesh distance metrics, per-lesion detection, fairness slices, and a confidence interval on every number.**
+
+[![PyPI](https://img.shields.io/pypi/v/segauge.svg)](https://pypi.org/project/segauge/)
+[![Python](https://img.shields.io/pypi/pyversions/segauge.svg)](https://pypi.org/project/segauge/)
+[![License](https://img.shields.io/badge/license-Apache--2.0-blue.svg)](LICENSE)
+
+segauge computes Dice, IoU, Hausdorff distance (HD), **HD95**, average symmetric surface distance (**ASSD**), **Normalized Surface Dice (NSD)**, and **per-lesion detection F1** for 3D medical image segmentation, and puts a **bootstrap confidence interval** on every number. It reads **NIfTI, DICOM-SEG, RTSTRUCT, and NumPy** directly, so you can evaluate the output of nnU-Net, MONAI, or TotalSegmentator without a lossy conversion.
+
+It exists because the metrics you report should be the metrics that are true, and today they often aren't.
+
+```bash
+pip install segauge
+```
+
+```python
+import segauge as sg
+
+result = sg.evaluate([
+    sg.Case("patient_001", pred="pred.nii.gz", gt="gt.nii.gz", metadata={"scanner": "siemens"}),
+    sg.Case("patient_002", pred="pred2.dcm",   gt="gt2.dcm",   metadata={"scanner": "ge"}),
+])
+
+print(result.summary())          # every metric with a 95% CI
+print(result.by_subgroup("scanner"))   # where does the model quietly fail?
+result.to_html("report.html")    # one self-contained report
+```
+
+Or from the command line:
+
+```bash
+segauge eval --pred preds/ --gt labels/ --metadata cases.csv --report report.html
+```
+
+## Why not just use MONAI or Metrics Reloaded?
+
+segauge bundles four things no incumbent provides together: a **confidence interval** on every metric, **per-lesion detection**, **subgroup/fairness slicing**, and **native DICOM-SEG/RTSTRUCT** input.
+
+It also computes distance metrics on a surface mesh rather than the voxel grid: it extracts the object surface with marching cubes **at true voxel spacing** and integrates over the surface, following the [MeshMetrics](https://arxiv.org/abs/2509.05670) method (Podobnik & Vrtovec, 2025). In our own benchmark (`benchmarks/mesh_vs_grid.py`) this measurably reduces **HD95** error under anisotropic, thick-slice spacing, where grid rasterization hurts most; for mean distance (ASSD) on smooth shapes the two methods are comparable. A formal validation suite against the MeshMetrics reference is in progress; we report what the benchmark shows, not more.
+
+| | segauge | MONAI | Metrics Reloaded | seg-metrics | DeepMind surface-distance |
+|---|:---:|:---:|:---:|:---:|:---:|
+| Surface-mesh distances (vs voxel grid) | ✅ | ❌ | ❌ | ❌ | ❌ |
+| Confidence intervals | ✅ | ❌ | ❌ | ❌ | ❌ |
+| Per-lesion detection F1 | ✅ | partial | ✅ | ❌ | ❌ |
+| Subgroup / fairness slicing | ✅ | ❌ | ❌ | ❌ | ❌ |
+| DICOM-SEG / RTSTRUCT native | ✅ | partial | ❌ | ❌ | ❌ |
+| `pip install` | ✅ | ✅ | ❌ | ✅ | ✅ |
+
+## Metrics
+
+- **Overlap:** Dice (DSC), IoU (Jaccard) — exact, integer-counted.
+- **Surface distance:** Hausdorff (HD), HD95, ASSD, MASD, Normalized Surface Dice (NSD) — mesh-based, spacing-aware, area-weighted.
+- **Per-lesion detection:** precision, recall, F1 via connected-component matching, so you can answer "did it find the tumor?" not just "how much voxel overlap?"
+- Every aggregate carries a deterministic **bootstrap confidence interval**.
+
+## Inputs
+
+NIfTI (`.nii`, `.nii.gz`), DICOM-SEG, RTSTRUCT, NumPy arrays, and `.npy`. Voxel spacing is read from the file header and used for the distance metrics, so a segmentation from a clinical pipeline is evaluated as-is.
+
+## FAQ
+
+**How do I compute HD95 correctly in Python?** `segauge.surface_metrics(pred, gt, spacing)` returns HD, HD95, ASSD, MASD, and NSD computed on the surface mesh, not the voxel grid.
+
+**Can it evaluate DICOM-SEG / RTSTRUCT directly?** Yes. `pip install segauge[dicom]` and pass a `.dcm` SEG, or use `segauge.load_rtstruct(series_dir, rtstruct, roi_name)`.
+
+**Does it work with nnU-Net / TotalSegmentator / MONAI outputs?** Yes. Point it at the NIfTI (or DICOM) they produce.
+
+**Why a confidence interval?** A Dice of 0.85 on 12 cases is not the same claim as 0.85 on 1200. segauge makes the difference visible.
+
+**2D images?** Overlap and detection metrics work in any dimension; surface-distance metrics are 3D in v0.1 (2D is planned for v0.2).
+
+## Status
+
+Pre-release (`0.1.0.dev`), built in the open. segauge is an evaluation tool for developers and researchers. It is **not a medical device** and produces no diagnosis.
+
+## References
+
+- Maier-Hein, Reinke et al. *Metrics Reloaded.* Nature Methods (2024). https://www.nature.com/articles/s41592-023-02151-z
+- Podobnik & Vrtovec. *MeshMetrics.* arXiv:2509.05670 (2025). https://arxiv.org/abs/2509.05670
+
+## License
+
+Apache-2.0.

segauge-0.1.0/benchmarks/mesh_vs_grid.py

@@ -0,0 +1,65 @@
+"""Mesh-correct vs grid-based surface distance: the thesis, with numbers.
+
+Grid-based tools (MONAI, seg-metrics, pymia, MedPy, DeepMind surface-distance)
+measure distances between voxel surfaces. segauge measures them on a marching-
+cubes surface mesh at true spacing. On geometry with a *known* answer, the gap
+shows up, and it widens with anisotropic spacing, exactly where it matters
+clinically (thick-slice CT/MR).
+
+Run:  uv run python benchmarks/mesh_vs_grid.py
+"""
+
+from __future__ import annotations
+
+import numpy as np
+from scipy import ndimage
+
+from segauge.metrics.distance import surface_metrics
+
+
+def sphere(shape, center, radius, spacing):
+    grids = np.ogrid[tuple(slice(0, s) for s in shape)]
+    dist2 = sum(
+        ((g - c) * sp) ** 2 for g, c, sp in zip(grids, center, spacing, strict=True)
+    )
+    return dist2 <= radius**2
+
+
+def _surface(mask):
+    return mask & ~ndimage.binary_erosion(mask)
+
+
+def grid_metrics(pred, gt, spacing, tol):
+    """Classic grid surface-distance, the way incumbent tools compute it."""
+    sa, sb = _surface(pred), _surface(gt)
+    dt_to_b = ndimage.distance_transform_edt(~sb, sampling=spacing)
+    dt_to_a = ndimage.distance_transform_edt(~sa, sampling=spacing)
+    d_ab, d_ba = dt_to_b[sa], dt_to_a[sb]
+    allv = np.concatenate([d_ab, d_ba])
+    nsd = ((d_ab <= tol).sum() + (d_ba <= tol).sum()) / (len(d_ab) + len(d_ba))
+    return {
+        "hd95": float(np.percentile(allv, 95)),
+        "assd": float(allv.mean()),
+        "nsd": float(nsd),
+    }
+
+
+def run(label, shape, spacing, r_inner, r_outer):
+    center = tuple(s / 2 for s in shape)
+    inner = sphere(shape, center, r_inner, spacing)
+    outer = sphere(shape, center, r_outer, spacing)
+    truth = r_outer - r_inner  # true symmetric surface distance, in physical units
+    mesh = surface_metrics(inner, outer, spacing=spacing, nsd_tolerance=1.0)
+    grid = grid_metrics(inner, outer, spacing, tol=1.0)
+    print(f"\n{label}  (spacing {spacing}, true ASSD = {truth:.1f})")
+    print(f"  {'metric':6} {'mesh':>10} {'grid':>10} {'truth':>10}")
+    for m, t in [("assd", truth), ("hd95", truth)]:
+        print(f"  {m:6} {mesh[m]:>10.3f} {grid[m]:>10.3f} {t:>10.3f}")
+    err_mesh = abs(mesh["assd"] - truth)
+    err_grid = abs(grid["assd"] - truth)
+    print(f"  ASSD abs error:  mesh {err_mesh:.3f}  vs  grid {err_grid:.3f}")
+
+
+if __name__ == "__main__":
+    run("isotropic 1mm", (90, 90, 90), (1.0, 1.0, 1.0), 30, 33)
+    run("anisotropic 1x1x3mm", (90, 90, 34), (1.0, 1.0, 3.0), 30, 33)

segauge-0.1.0/docs/api.md

@@ -0,0 +1,23 @@
+# API reference
+
+## Top-level
+
+::: segauge.evaluate
+
+::: segauge.Case
+
+::: segauge.surface_metrics
+
+::: segauge.dice
+
+::: segauge.iou
+
+::: segauge.detection_scores
+
+::: segauge.bootstrap_ci
+
+## Loaders
+
+::: segauge.load_mask
+
+::: segauge.load_rtstruct

segauge-0.1.0/docs/compute-hd95-in-python.md

@@ -0,0 +1,37 @@
+# How to compute HD95 (95th-percentile Hausdorff distance) in Python
+
+**HD95** is the 95th percentile of the symmetric surface distances between a predicted and a ground-truth segmentation. It is preferred over the maximum Hausdorff distance (HD) because it is far less sensitive to a single outlier voxel, which is why Metrics Reloaded recommends it for most segmentation tasks.
+
+With segauge:
+
+```python
+import segauge as sg
+
+metrics = sg.surface_metrics(pred_mask, gt_mask, spacing=(1.0, 1.0, 3.0))
+print(metrics["hd95"])   # 95th-percentile Hausdorff distance, in mm
+print(metrics["hd"])     # maximum Hausdorff distance
+print(metrics["assd"])   # average symmetric surface distance
+```
+
+`pred_mask` and `gt_mask` are boolean (or 0/1) 3D NumPy arrays. `spacing` is the physical voxel size in the same axis order as the arrays; it matters, because HD95 is a physical distance.
+
+## Why segauge's HD95 is different
+
+Most libraries compute Hausdorff distance on the **voxel grid**: they take the surface voxels and measure distances between voxel centres. That bakes in discretization error, which is largest for thick-slice (anisotropic) data.
+
+segauge extracts the object surface with marching cubes at the true voxel spacing and measures distances over that surface, following the [MeshMetrics](https://arxiv.org/abs/2509.05670) method. In benchmarking on data with a known answer, this reduces HD95 error under anisotropic spacing.
+
+## From files (NIfTI / DICOM-SEG)
+
+```python
+import segauge as sg
+
+result = sg.evaluate([sg.Case("p1", pred="pred.nii.gz", gt="gt.nii.gz")])
+print(result.summary()["hd95"])   # HD95 with a bootstrap confidence interval
+```
+
+## Confidence intervals
+
+When you evaluate more than one case, every metric (including HD95) comes back with a [bootstrap confidence interval](confidence-intervals.md), so you can report HD95 honestly.
+
+See also: [Normalized Surface Dice](normalized-surface-dice-nsd.md), [the API reference](api.md).

segauge-0.1.0/docs/confidence-intervals.md

@@ -0,0 +1,39 @@
+# Segmentation Dice (and HD95) with confidence intervals
+
+A mean Dice of 0.85 over 12 cases is a very different claim from 0.85 over 1200 cases, but papers usually report just the number. segauge attaches a **bootstrap confidence interval** to every aggregate metric so the reader can tell the difference, and so you can report results honestly.
+
+```python
+import segauge as sg
+
+result = sg.evaluate([
+    sg.Case("p1", pred="p1.nii.gz", gt="g1.nii.gz"),
+    sg.Case("p2", pred="p2.nii.gz", gt="g2.nii.gz"),
+    # ...
+])
+
+for name, est in result.summary().items():
+    print(f"{name}: {est.value:.3f} [{est.ci_low:.3f}, {est.ci_high:.3f}]")
+# dice: 0.910 [0.882, 0.937]
+# hd95: 3.20 [2.10, 4.80]
+# ...
+```
+
+Every metric, Dice and IoU and HD95 and ASSD and NSD and detection F1, comes with a 95% interval. The interval is **deterministic** given a seed, because a tool you trust to report results must be reproducible.
+
+## Confidence interval for your own values
+
+```python
+import segauge as sg
+
+per_case_dice = [0.88, 0.91, 0.79, 0.93, 0.85]
+est = sg.bootstrap_ci(per_case_dice)
+print(est)   # 0.872 [0.812, 0.918]
+```
+
+## Configure it
+
+```python
+result = sg.evaluate(cases, confidence=0.95, n_resamples=2000, seed=0)
+```
+
+See also: [Per-lesion detection F1](per-lesion-detection-f1.md), [the API reference](api.md).