write-imaging-factsheet

Write Imaging Factsheet

1. For each image (current and prior, if supplied), call the medimageparse MCP tools (VS Code may list them as mcp_medimageparse_<tool>): list_models, then segment with model mip3d-mri-brain-lesion and its tumor-components.prompts verbatim, in order. If the tools are not available or if the run fails, stop and tell the user to check the medimageparse server in the tool picker. Do not run the server's Python code, write your own segmentation or measurements, or write the factsheet with a script.
2. Call clean_segmentation on each returned mask_path. Keep the original masks, clean masks, lesion-instance masks, and sidecars.
3. Call describe_lesions with the current clean_path, passing prior_clean_path when available. Keep all descriptor JSON files.
4. Fill template/factsheet.md from the returned descriptors only. Replace ID placeholders with actual IDs; repeat lesion, pair, total, change, and active-limitation lines as needed. Copy numbers, units, categories, booleans, and reasons verbatim. Cite descriptor IDs in brackets on every fact line. Copy frontmatter from inputs and conventions; use null for a missing prior. Omit sections with no descriptors. For unavailable values, copy null and the reason; never invent a value. Do not compute, round, diagnose, assign anatomy or side, or imply lesion matching. Save <mask-stem>.factsheet.md beside <mask-stem>.descriptors.json (replace .descriptors.json with .factsheet.md). In the Images section, link the current scan's review images by file name (they sit beside the factsheet); drop the measurements line if describe_lesions returned none for it. Reply with the factsheet, then end with each scan's audit-log folder (audit_dir from segment), current first; the current folder holds the factsheet and images.