Repository Information

• Repository File Structure
• Documentation Guide
  • Building the documentation on your workstation
  • Previewing and checking the site
  • Checking a pull request’s documentation
  • Source files and publishing
  • Writing portable links
  • Keeping documentation current
  • Docstring extraction
• pymedphys Release Procedure
  • Checklist
  • Choose the versions
  • Prepare the release pull request
  • Publish the release
  • Start the next development version
  • Check the published files yourself
  • Recover from failures
  • Appendix: publishing setup and access
• PyMedPhys CI/CD Workflows
  • Overview
  • Workflow Architecture
  • Workflow Structure
  • Composite Actions
  • PR Workflow
  • Merge Queue Workflow
  • Main Branch Workflow
  • Required Secrets
  • Environments
  • Read the Docs
  • Required checks and pull request reviews
  • Labels for Manual Triggers
  • Testing Workflows Locally
  • Security Considerations
  • Version Compatibility
• Reviewing automated dependency update PRs
  • What creates these PRs?
  • What validation should run before the PR opens?
  • Fast review path
  • What the reviewer should do
  • Important note about the PR body
  • When to slow down
  • Manual commands for a cautious reviewer
  • Merge checklist
  • What if no dependency update PR appeared?
• Lazy imports and optional dependencies
  • Base install and extras
  • How lazy imports work
  • What a user sees when a package is missing
  • Rules for contributors
  • How it is checked
  • Known gap
• DICOM de-identification design
  • Scope
  • Terminology and claims
  • Standards basis
  • Requirements register
  • Architecture
  • Presets
  • Decisions
  • Roadmap
• DICOM de-identification requirements-to-tests matrix
  • Summary
  • Requirements
  • Exclusions
  • Remaining work
  • Details
• Run the MIDI-B de-identification benchmark
  • 1. Install Git and uv
  • 2. Get PyMedPhys with its DICOM dependencies and decoders
  • 3. Download MIDI-B
  • 4. Run the benchmark
  • 5. What it writes
  • 6. Share the results
  • 7. Optional: score the same release with the NCI script
  • 8. The Test subset
• Validate de-identified DICOM with independent validators
  • The validators
  • How the comparison works
  • Run it on the synthetic corpus
  • Run it on a release of your own
  • Read a report
• Check the face block against HaN-Seg
  • The data set
  • Run the check
  • What the counts show
  • Reading the results
  • Limits
  • Check DICOM exports on your own machine
• Check the text detector against HaN-Seg
  • The data set
  • Run the check
  • What the counts show
• DICOM coordinate validation
  • Independence of the checks
  • Evidence and its limits
  • Retirement of downloaded test fixtures
  • Remaining limitations
  • Reproducing the committed checks
• DICOM coordinates and gamma, illustrated
  • Start with a coordinate–dose pair
  • Patient directions and file terminology
  • 1. From stored pixels to patient coordinates
  • 2. The current conversion places every voxel correctly
  • 3. Edge cases
  • 4. How gamma treats axis order
  • 5. Plotting gamma after the reordering
  • 6. Checking an earlier calculation
  • 7. Remaining limitations
  • Appendix A. Why descending evaluation axes failed before
  • Further reading: gamma performance
• Faster gamma calculations: a reproducible benchmark
  • 1. What is held constant?
  • 2. Reproducible inputs and measurements
  • 3. Results: complete gamma-call runtime
  • 4. Does the improvement recur across rounds?
  • 5. Is the numerical answer unchanged?
  • 6. Why can the same calculation take less time?
  • 7. Repeat the benchmark on your computer
• Compare gamma performance across PyMedPhys versions
  • Prepare a comparison
  • Revisit the SciPy and PyMedPhys interpolators
  • Choose and freeze the workload
  • Execute and resume
  • Measurement protocol
  • Interpret and retain the report
  • Environment setup
• Gamma performance: a two-hour workstation audit
  • Start
  • A fixed scope that fits the budget
  • The workloads and their interpretation
  • Timing, numerical checks and ratios
  • Progress, stopping, resuming and upload
• Gamma performance: a completed workstation audit
  • Absolute waiting times for the larger volumes
  • How much improvement comes from this PR?
  • A useful linear model within a specified workload
  • Does the model predict the separate volumes?
  • Why voxel count alone cannot determine runtime
  • Coverage, numerical checks and provenance
  • Inspect or reproduce the evidence
• Gamma scaling: recorded workstation evidence
  • What the completed measurements show
  • PyMedPhys versus SciPy
  • Coverage and numerical checks
  • Scaling curves
  • Source and reproducibility