菜单
Prototype scaffold for sliding-window volume inference interfaces. This skill demonstrates window generation, batching, aggregation, and reporting, but it does not provide production-ready checkpoint-to-model loading.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Scan medical imaging directories and generate structured JSON or CSV manifests with file-level metadata for DICOM and NIfTI assets.
Validates DICOM series consistency by checking SeriesInstanceUID, ImageOrientationPatient, slice spacing uniformity, and detecting missing slices.
Normalizes medical image intensities using various methods (z-score, min-max, percentile, window-level). Essential preprocessing step for consistent analysis across different scanners and protocols.
Generate research-oriented HTML or Markdown case reports for volumetric medical images, including summary statistics and slice montages.
Resample NIfTI medical images to target spacing with configurable interpolation methods and optional anti-aliasing for downsampling.
Generates a complete MONAI-based training scaffold for medical image segmentation or classification tasks. Creates training scripts, inference scripts, configuration files, data loading pipelines, and model definitions.
| name | sliding-window-inference-runner |
| description | Prototype scaffold for sliding-window volume inference interfaces. This skill demonstrates window generation, batching, aggregation, and reporting, but it does not provide production-ready checkpoint-to-model loading. |
| version | 1.0.0 |
| author | Medical AI Skills Team |
| license | MIT |
| tags | ["inference","segmentation","sliding-window","scaffold","pytorch"] |
| maturity | scaffold |
| category | segmentation-ml |
| requires_extras | ["core","ml"] |
This skill is a scaffold for experimenting with sliding-window inference interfaces on volumetric data. It is useful for testing data tiling, Gaussian weighting, aggregation, and CLI/report plumbing.
It is not a production-ready checkpoint-driven inference engine. External users should not treat its outputs as real model predictions.
pip install .[ml]
Dependencies:
python scripts/main.py --input IMAGE_PATH --model MODEL_PATH --output OUTPUT_PATH [OPTIONS]
| Argument | Required | Description |
|---|---|---|
--input | Yes | Input image path |
--model | Yes | Placeholder checkpoint path used to infer scaffold metadata |
--output | Yes | Output segmentation path |
| Option | Default | Description |
|---|---|---|
--window-size | 96 | Sliding window size |
--overlap | 0.25 | Window overlap ratio |
--batch-size | 1 | Batch size |
--device | cpu | Device to run the scaffold on |
--report | auto | Optional JSON report path |
--version | - | Show version and exit |
--help | - | Show help and exit |
Input is a 2D or 3D image plus a checkpoint-like file path. The current implementation does not reconstruct the original model architecture from the checkpoint.
{
"success": true,
"result": {
"output_path": "segmentation.nii.gz",
"report_path": "segmentation.json",
"maturity": "scaffold",
"warning": "Output was produced by a scaffold pipeline, not real checkpoint-driven inference."
},
"metadata": {
"timestamp": "2024-01-15T10:30:00Z",
"duration_ms": 500,
"version": "1.0.0"
}
}
| Code | Meaning |
|---|---|
| 0 | Success |
| 1 | Validation error |
| 2 | Runtime error |
| 3 | Usage error |
python scripts/main.py --input image.npy --model dummy_model.pth --output scaffold_mask.npy
python scripts/main.py --input image.nii.gz --model dummy_model.pth --output scaffold_mask.nii.gz --report scaffold_report.json
For research use only. This skill is scaffold-only and does not provide production-ready medical model inference.