<< All versions

Skill v1.0.0

currentTrusted Publisher100/100
microsoft/healthcareai-examples/ground-zeroshot
──Details
PublishedSeptember 26, 2026 at 12:19 AM
Content Hashsha256:653bade9e933998b...
Git SHA
──Files
Files (1 file, 2.5 KB)
SKILL.md2.5 KBactive
SKILL.md · 57 lines · 2.5 KB

version: "1.0.0" name: ground-zeroshot description: Ground CXR findings using MI2 zero-shot classification. Returns per-finding probabilities bucketed into positive/possible for the agent to use when writing the report. argument-hint: |

  • image_path: required file path or http(s) URL to the CXR image (PNG or JPEG)
  • extra_labels: optional list of additional label strings to score alongside the default set

ground-zeroshot

Calls the MI2 zero-shot classifier and returns bucketed per-finding probabilities. Does not write a report — the agent produces the report using the write-report skill after grounding.

Budget

  • Tool calls: 1 — exactly one mcp_medimageinsig_zeroshot_classify call per image.
  • Do not retry on success.

How to use

  1. Resolve the image. If image_path is an http(s) URL, pass it through to the tool unchanged — do NOT download it, curl it, or turn it into a local file; the MCP server fetches URLs itself. Otherwise it is a local path: the MCP tool runs in a different working directory, so if it is not absolute, resolve it via realpath; if it can't be resolved, ask the user.
  1. Assemble the label list. Read the canonical label list from below and use those strings verbatim. Append any caller-provided extra_labels.
  1. Call the classifier. Call mcp_medimageinsig_zeroshot_classify with file_path = absolute path and labels = assembled list. One call per image.
  1. Bucket the scores. Use these cutoffs:
  • ≥ 0.20 → positive
  • 0.08 – 0.20 → possible
  • < 0.08 → absent (omit)
  1. Return the findings. Return the positive and possible labels with their raw scores. Do not write a report. The agent decides what to do next.

NIH ChestX-ray14 labels (MI2 format)

Pass these strings as the labels argument to zeroshot_classify. Format is "<modality> <body_part> <view> <condition>" — MI2 is sensitive to this structure, do not reword.

x-ray chest anteroposterior Atelectasis
x-ray chest anteroposterior Cardiomegaly
x-ray chest anteroposterior Consolidation
x-ray chest anteroposterior Edema
x-ray chest anteroposterior Effusion
x-ray chest anteroposterior Emphysema
x-ray chest anteroposterior Fibrosis
x-ray chest anteroposterior Hernia
x-ray chest anteroposterior Infiltration
x-ray chest anteroposterior Mass
x-ray chest anteroposterior Nodule
x-ray chest anteroposterior Pleural_Thickening
x-ray chest anteroposterior Pneumonia
x-ray chest anteroposterior Pneumothorax
All versions