habit-radiomics

name	habit-radiomics
description	Extract traditional PyRadiomics features (firstorder, shape, GLCM, GLRLM, GLSZM, NGTDM, GLDM) from medical images at the whole-tumor ROI level — no habitat segmentation. Use when the user wants classical radiomics features without doing habitat analysis. Triggers on "传统影像组学", "PyRadiomics", "radiomics features", "shape features", "GLCM". Runs `habit radiomics`.

HABIT Traditional Radiomics

Extract whole-tumor PyRadiomics features (no habitat clustering involved). The classical radiomics workflow — one row per subject, one feature per column.

If the user wants habitat-based features, redirect to habit-feature-extraction.

CLI

habit radiomics --config <config_traditional_radiomics.yaml>

Required Information

Field	Stop if missing
paths.params_file	optional — omit for bundled parameter.yaml preset; override with path or @preset:roi
paths.images_folder	yes
paths.out_dir	yes
processing.process_image_types	yes — modality folder names

PyRadiomics parameter file

Three options for the params file:

File	Use
config/radiomics/parameter.yaml	generic full set
config/radiomics/parameter_basic.yaml	minimal (~70 features)
config/radiomics/parameter_with_filters.yaml	full with LoG+Wavelet (~1500 features)

Choosing guide: references/parameter_choice_guide.md.

If the user has no params file, omit paths.params_file (bundled default) or use parameter_basic.yaml as a starting point for a custom file.

Standard config

paths:
  # params_file omitted → bundled parameter.yaml preset
  images_folder: ./data/preprocessed_images
  out_dir: ./results/radiomics

processing:
  n_processes: 4
  save_every_n_files: 5
  process_image_types:
    - T1
    - T2
    - DWI
    - ADC

export:
  export_by_image_type: true       # one CSV per modality
  export_combined: true            # merged CSV with all modalities (prefixed)
  export_format: csv               # csv | json | pickle
  add_timestamp: true

logging:
  level: INFO
  console_output: true
  file_output: true

Reference templates

Config index: skills/CONFIG_SOURCES.md.

File	Use
config/radiomics/config_traditional_radiomics.yaml	scaffold
config/radiomics/parameter.yaml	starter PyRadiomics params
references/parameter_choice_guide.md	how to pick filters and feature classes

Full annotated reference: config/radiomics/config_traditional_radiomics.yaml.

Validate output (after run)

python skills/habit-feature-extraction/scripts/inspect_feature_csv.py \
  <out_dir>/radiomics_features_combined_*.csv --subject-id-col subject_id

(The same CSV inspector used for habitat features works here too.)

Output files

With default settings (add_timestamp: true, both export options on):

out_dir/
├── radiomics_features_T1_<timestamp>.csv      # per-modality
├── radiomics_features_T2_<timestamp>.csv
├── radiomics_features_DWI_<timestamp>.csv
├── radiomics_features_ADC_<timestamp>.csv
├── radiomics_features_combined_<timestamp>.csv  # all modalities merged
└── extraction.log

The combined CSV is the typical input for habit model.

Differences from habit extract

Aspect	habit radiomics	habit extract
Operates on	Whole-tumor ROI	Habitat sub-regions
Requires	Just images + masks	Habitat maps too
Output features	Per modality	Per habitat + traditional + MSI + ITH
Use case	Classical radiomics	Habitat-based study

If the user has habitat maps and wants comprehensive features, prefer habit extract (which can include traditional whole-tumor features as one of its options).

Common pitfalls

1. process_image_types mismatch — names must exactly match folder names. Case-sensitive.
2. PyRadiomics fails on small ROIs — ROIs <30 voxels can fail GLCM/GLRLM. Increase binWidth or check ROI size.
3. Resampling mismatch — resampledPixelSpacing in parameter.yaml may resample again. Comment out if already resampled in preprocessing.
4. Memory blow-up with Wavelet + LoG — each filter multiplies features ~8-10×. Start with Original: {} only for testing.
5. Mask label value — default is label=1. If multi-label mask, specify label: 1 in parameter.yaml.

For more, see habit-troubleshoot/references/errors_extraction.md.

Verification

After running:

• Open one CSV — should have ~100-1500 columns depending on params
• First column = subject ID
• No NaN columns (NaN = failed extraction → check log)

Next step

After radiomics extraction, proceed to habit-machine-learning to train models.