DICOM Reader

DICOM Reader (Split Tools)

Author: langgenius
Version: 0.0.2
Type: Plugin (multiple tools)

Overview

This plugin now provides focused DICOM tools, split from the original monolithic reader. Each tool does one job well and keeps outputs compact for LLMs.

Refactor highlights:

• Extracted common helpers into tools/dicom_reader/tools/_utils.py to reduce duplication.
• Unified preview image generation with a consistent size cap and resizing policy.
• Standardized frame selection, type coercion, and simple parsing utilities across tools.

Available tools and typical use cases:

• dicom_metadata

  • Purpose: Load patient/study/series/image tags and optional extra tags
  • Example: ds = pydicom.dcmread(file)

• dicom_pixels

  • Purpose: Summarize pixel array (shape/dtype/stats) with optional preview
  • Example: img = ds.pixel_array

• dicom_multiframe

  • Purpose: Read multi-frame CT/MRI, report frame count, preview a frame
  • Example: ds.NumberOfFrames

• dicom_hu_correction

  • Purpose: Apply RescaleSlope/Intercept; return stats and a preview
  • Example: hu = img * slope + intercept

• dicom_spatial

  • Purpose: Extract pixel spacing, slice thickness, orientation/position; compute voxel size/volume

• dicom_pixel_ops

  • Purpose: Add/subtract, normalize, clip, contrast stretch, box blur; returns preview

• dicom_stats

  • Purpose: Mean/variance/std and histogram on full image or rectangular ROI

• dicom_volume

  • Purpose: Threshold-based mask with spacing to estimate volume (mm³) and mean intensity

• dicom_threshold_mask

  • Purpose: Build a binary mask from thresholds and return overlay preview

• dicom_roi

  • Purpose: Compute stats inside a rectangular ROI; optional outlined preview

• dicom_model_input

  • Purpose: Convert to standard model shape [1, C, H, W] (or NHWC); optional normalization and preview

The original combined tool (tools/dicom_reader.yaml) remains for backwards compatibility.

Common Parameters

• dicom_file (file, required): DICOM Part 10 (.dcm) file to process
• frame_index (number): 0-based frame index for multi-frame data (tools that use frames)
• max_preview_edge (number): Max preview edge (32–1024) for tools that output images

Large File Strategy

• Where possible, headers are parsed with stop_before_pixels to keep responses small.
• Tools that operate on pixel data avoid returning full arrays; instead they report stats and attach small PNG previews (soft-capped to ~8 MB per image).

Shared Utilities

• as_bool/as_int/as_float: consistent parameter parsing
• select_frame: robust multi-frame indexing across shapes
• ensure_hw_or_hwc and to_uint8_minmax: reliable image normalization
• make_preview_png_bytes: preview generation with edge clamping and byte-size cap
• Metadata helpers (file_stats, collect_named_fields, parse_tag_keyword, etc.) for compact, human-friendly output

Notes

• No external credentials required.
• Outputs are JSON and small binary attachments (PNG) only; no base64 in JSON payloads.