Skip to main content

Preprocess

This document describes the preprocessing pipeline for CODEX images in Aegle. The preprocessing consists of two main stages: Full Image Preprocessing and Patched Image Preprocessing, as implemented in aegle/pipeline.py under the run_pipeline function.

Key outputs include:

A. Full Image Preprocessing​

This stage prepares the full CODEX image for downstream patch-level processing.

Step A1: Initialize CodexImage​

  • Constructs a CodexImage object using configuration parameters and command-line arguments.
  • Loads the image and its corresponding antibody metadata.
  • If enabled in the configuration (data.generate_channel_stats: true), calculates statistics across channels.
codex_image = CodexImage(config, args)
codex_image.calculate_channel_statistics()

Step A2: Extract Target Channels​

  • Uses configuration to extract biologically relevant channels (e.g. membrane or nucleus markers).
codex_image.extract_target_channels()

Optional: Visualize Whole Sample​

  • Saves RGB visualizations of extracted channels to the output directory if visualization.visualize_whole_sample is set to true.
save_image_rgb(codex_image.extended_extracted_channel_image, ...)
save_image_rgb(codex_image.extracted_channel_image, ...)

B. Patched Image Preprocessing​

This stage slices the image into overlapping patches and optionally applies perturbations for robustness testing.

Step B1: Extend Image​

  • Pads the original image to ensure that all patches are fully contained within the image bounds.
codex_image.extend_image()

Step B2: Initialize CodexPatches and Save Patches​

  • Initializes a CodexPatches object using the extended CodexImage, then generates and saves patches.
  • Patch metadata is also saved.
codex_patches = CodexPatches(codex_image, config, args)
codex_patches.save_patches(...)
codex_patches.save_metadata()

Optional: Add Disruptions​

  • If configured, adds synthetic noise or perturbations to patches for testing robustness.
  • Disruption parameters include type and level, and the disrupted patches can be optionally saved and visualized.
disruption_type = disruption_config.get("type")
disruption_level = disruption_config.get("level", 1)
codex_patches.add_disruptions(disruption_type, disruption_level)
codex_patches.save_disrupted_patches()

Optional: Visualize Patches​

  • RGB visualizations of either the original or disrupted patches are saved if visualization.visualize_patches is set to true.
save_patches_rgb(codex_patches.extracted_channel_patches, ...)

This preprocessing module prepares both the full and patched images for subsequent segmentation and analysis steps. All outputs are saved in a structured format for downstream reproducibility.