Skip to main content

qwen-mm-plugins-nifti

Inspect local NIfTI (.nii/.nii.gz) header metadata or render slices with shared intensity normalization and window presets. Use for checking volume geometry, NIfTI viewing, choosing slices or windows, and verifying effective display settings.

ソース情報

リポジトリ
QwenLM/Qwen-MM-Plugins
ソースの最終更新活動
2026年10月1日 18:26
検出された SKILL.md の言語
英語
スター
3,135
フォーク
209

インストール方法

デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。

ソースファイルを確認

インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。

SKILL.md を表示中

SKILL.md
ソースの指示 · 読み取り専用プレビュー
name
qwen-mm-plugins-nifti
description
Inspect local NIfTI (.nii/.nii.gz) header metadata or render slices with shared intensity normalization and window presets. Use for checking volume geometry, NIfTI viewing, choosing slices or windows, and verifying effective display settings.
# NIfTI Volume Inspection Use the `qwen-mm-plugins-nifti` MCP server's two tools according to the request: - `nifti_inspect`: read header metadata when the user asks about shape, dtype, spacing, units, affine, orientation, obliquity, or the number of 3D volumes. It returns JSON in a text block without scanning voxel intensities or rendering images. It cannot determine intensity ranges, modality, or anatomy from pixels. - `nifti_render_slices`: render images with configurable slice selection and intensity/window mapping. Call it directly for viewing; inspection is optional, not a prerequisite or a stateful initialization step. When available, prefer these tools for their respective NIfTI tasks. Respect an explicit request to use another tool. ## Read and verify Pass an absolute `.nii` or `.nii.gz` file path to either tool. Both read the source without modifying it. Rendering in the default native mode returns text and image blocks; if `QWEN_MM_NATIVE_MODE=0`, the framework's caption fallback can send rendered slices to the configured VL endpoint. With only `file_path`, `nifti_render_slices` selects three uniformly spaced interior slices along **source voxel axis 2**. Positions come from a five-point grid with its endpoints discarded, rounded to source indices. Small dimensions can resolve to fewer unique slices. Axis 2 is not necessarily anatomical axial/Z: check the reported affine, closest orientation codes, and obliquity. In-plane axes are reordered/flipped for display; no anatomical resampling is performed. For 4D inputs, rendering defaults to the first 3D volume. `volumes` selects 1-based volume pages, for example `"1,3"`; `max_volumes` caps this selection. The fourth dimension is not automatically interpreted as time. Spatial units follow the header; an unknown unit is explicitly reported as assumed mm. ## Select slices and intensity - Set `slice_axis` to 0, 1, or 2 and `num_slices` for evenly spaced slices. - For exact locations, set either `slice_indices` (0-based) or `slice_positions` (0–1 along the source axis). Either overrides `num_slices`; they cannot be combined. Duplicate resolved indices are returned once, in requested order. - `intensity_mode="auto_volume"` uses one shared P1–P99 range per selected 3D volume. `percentile_low` and `percentile_high` adjust the bounds. Large volumes use a deterministic grid approximation; the response identifies exact/sampled statistics and the effective range. Sampling statistics does not subsample the displayed planes. - `intensity_mode="window"` requires both `window_center` and `window_width`. - `intensity_mode="preset"` requires `window_preset`: `ct_brain`, `ct_soft_tissue`, `ct_lung`, or `ct_bone`. Presets assume HU-like CT values; NIfTI alone does not establish modality or calibrated intensity units. Use a preset when requested or supported by user-provided acquisition context, and distinguish a suggested window from the one actually applied. For metadata-only requests, report the relevant header values, preserving any unit assumption and the distinction between voxel axes and anatomical axes. Inspection does not choose slices or apply normalization; do not invent display settings for that response. The fourth dimension is not necessarily time. For rendered views, report the tool used, selected volume(s), source axis and resolved indices, intensity mode and effective bounds, and whether statistics were sampled. Mention any response-size truncation or assumed unit relevant to the request. Use the returned configuration rather than reconstructing it from requested arguments. If more slices are needed, request the omitted indices/volumes in another call. This tool provides visualization, not a clinical diagnosis.
GitHubで見る