segauge: medical image segmentation metrics in Python

segauge is a Python library for evaluating medical image segmentation. It computes Dice, IoU, Hausdorff distance (HD), HD95, average symmetric surface distance (ASSD), Normalized Surface Dice (NSD), and per-lesion detection F1, 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.

pip install segauge

import segauge as sg

result = sg.evaluate([
    sg.Case("patient_001", pred="pred.nii.gz", gt="gt.nii.gz", metadata={"scanner": "siemens"}),
])
print(result.summary())            # every metric with a 95% confidence interval
print(result.by_subgroup("scanner"))   # see where a model quietly fails
result.to_html("report.html")      # one self-contained report

What you can do with it

• Compute HD95 in Python — 95th-percentile Hausdorff distance, on a surface mesh at true voxel spacing.
• Normalized Surface Dice (NSD) — surface agreement within a clinical tolerance.
• Evaluate a DICOM-SEG — read a DICOM segmentation object directly.
• Dice with confidence intervals — because 0.85 on 12 cases ≠ 0.85 on 1200.
• Per-lesion detection F1 — "did it find the tumor?", not just voxel overlap.
• How segauge compares to MONAI and Metrics Reloaded.

Why segauge

It bundles four things no existing library provides together: confidence intervals on every metric, per-lesion detection, subgroup/fairness slicing, and native DICOM-SEG/RTSTRUCT input. Distance metrics are computed on a surface mesh at true voxel spacing rather than on the voxel grid, following the MeshMetrics method; in benchmarking this reduces HD95 error under anisotropic, thick-slice spacing. See the comparison.

segauge is an evaluation tool for developers and researchers. It is not a medical device.