spatialShrunkenCentroids Docker

A Dockerized R application for performing post-hoc feature selection on Mass Spectrometry Imaging (MSI) data using Cardinal::spatialShrunkenCentroids.

Features

• Post-hoc Feature Selection: Identifies specific m/z values that differentiate spatial clusters.
• Niivue & M2aia Compatible: Outputs uncompressed (raw) NRRD files that load perfectly in standard volumetric viewers.
• Cross-Version Support: Poly-compatible with Cardinal 3.4.x (Bioconductor containers) and 3.12.x+.
• Rich Metadata: Embeds analysis parameters (r, s, AIC, BIC, Sparsity) and timestamps directly into NRRD headers.

Usage

1. Build the Image

docker build -t spatial-shrunken-centroids .

2. Run Analysis

Mount your project directory to /data and provide paths relative to that mount.

docker run --rm -v "C:/path/to/data:/data" spatial-shrunken-centroids \
  --input /data/intensity_nrrds \
  --labels /data/mask.nrrd \
  --output /data/results.nrrd \
  --s 2 \
  --r 1

Arguments

• --input: Directory containing one .nrrd file per m/z (filenames should be numeric m/z values).
• --labels: Path to the segmentation mask .nrrd file (labels should be integers, 0 reserved for background).
• --output: Path for the predicted labels NRRD.
• --s: Sparsity threshold (higher values = more sparse feature selection).
• --r: Spatial smoothing radius.

Outputs

• results.nrrd: The predicted cluster labels (16-bit unsigned integer).
• probabilities_cluster_*.nrrd: Probability maps for each cluster (32-bit float).
• importance.csv: Table of features ranked by their importance to each cluster.
• NRRD Headers: Contains metadata for reproducibility:
  • Cardinal_SSC_r: Radius used.
  • Cardinal_SSC_s: Sparsity used.
  • Cardinal_SSC_Sparsity: Fraction of features removed.
  • Cardinal_SSC_AIC / BIC: Model quality metrics.

Input Format Requirements

• NRRD Filenames: Files in the --input directory should be named with their m/z value (e.g., 801.45.nrrd).
• Dimensions: All input NRRDs and the label mask must have identical X, Y (and Z) dimensions.
• Orientation: Assumes standard MSI orientation (X varies fastest).

Technical Implementation

This tool uses Cardinal::spatialShrunkenCentroids with "adaptive" weights. It is designed to mirror the interface of spatialKmeansDocker to facilitate seamless multi-stage pipelines.