MR-OOD: Visualization & Reporting Suite

This repository contains the tools and Jupyter Notebooks used to generate data statistics and produce visualizations for the MR Out-of-Distribution (OOD) anomaly detection project.

Features

• Data Diagnostics: Verify NIfTI consistency across different channel formats (1-channel vs. 2.5D).
• Statistical Reporting: Generate patient and slice distribution analysis for dataset versioning.
• Comparative Visualization: Standardized reporting of raw MR, ground truth, and model predictions.

---

File Overview

Python Scripts

• visualize.py: A command-line tool designed to generate standardized 3-panel comparison plots.

Jupyter Notebooks

• channel_test.ipynb: Diagnostic tool to verify NIfTI image consistency and distinguish between 2.5D and 1-channel data formats.
• dataset_stats.ipynb: Generates bar charts comparing patient and slice distributions across ID/OOD splits and dataset versions.
• mask_refinement.ipynb: Visualization of the mask refinement procedure and morphological operations.
• discussion_visualization.ipynb: Research notebook for testing Ground Truth masks against predictions from multiple anomaly detection models.

---

Usage: Visualization Script

The visualize.py script generates side-by-side comparisons of the MR, the ground truth mask, and the predicted model mask. It produces a 1x3 grid for a specific slice (if provided) or for all slices in a patient volume containing masks.

Command Example

python visualize.py \
  --mr_path /path_to_mr/1PA118/mr.nii.gz \
  --predicted_mask_path /path_to_prediciton/PA118.nii.gz \
  --ground_truth_path /path_to_ground_truth/PA118.nii.gz \
  --output_directory ./results_visualization \
  --model_name "PatchCore" \
  --saving_format png \
  --patient_slice 47

Example Vizualization