---

name: outputs-and-statistics description: "Consume TotalSegmentator masks, run reports, statistics, radiomics, probabilities, multilabel files, and combined-mask outputs." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

TotalSegmentator Outputs and Statistics

Use this sub-skill when a segmentation has already been run and the next task is to validate, summarize, post-process, or combine its outputs. For running a segmentation, route to segmentation-workflows; for task and class discovery, route to capability-discovery; for DICOM output mechanics, route to dicom-and-formats.

What This Covers

• Validate --report JSON manifests without scraping stdout.
• Read statistics.json, statistics_radiomics.json, and --statistics_extra metrics.
• Understand per-class masks, --ml multilabel outputs, label maps, and --save_probabilities side outputs.
• Build safe totalseg_combine_masks commands for lung, ribs, vertebrae, heart, pelvis, and body masks.
• Diagnose common output/report/statistics failures after a run completes.

Quick Paths

• Run report schema and validation: references/reports-and-statistics.md
• Output format behavior and post-processing commands: references/output-formats.md
• Troubleshooting matrix: references/troubleshooting.md
• Summarize or enforce a report: python scripts/inspect_run_report.py --report seg/run_report.json --require-task total --require-class liver
• Inspect a multilabel label map: python scripts/inspect_multilabel_header.py seg.nii.gz --require-label spleen
• Build a combine command: python scripts/combine_masks_command.py --input seg/ --output lung.nii.gz --mode lung

Operating Rules

• Treat run_report.json and statistics.json as the automation contract; do not parse human progress text.
• Use report.classes to interpret class names, especially after --roi_subset filters the run.
• Expect report.output_files to be populated only when output is a directory of NIfTI files; it can be empty for single-file multilabel or DICOM-style outputs.
• Do not rerun segmentation to combine masks; use existing per-class masks or a multilabel NIfTI with totalseg_combine_masks.
• If a class or task name is unknown, route to capability-discovery and query totalseg_info instead of hard-coding names.