LesionView Online

A browser-based web application for visualizing and analyzing longitudinal MS lesion data. Built with React Native for Web (Expo).

Features

• NIfTI File Loading: Load FLAIR images and lesion masks directly in the browser (no server required).
• Dual Timepoint Comparison: Compare baseline and follow-up scans side-by-side.
• Lesion Matching & Classification: Automatically match lesions across timepoints using graph-based connectivity.
• Interactive 3D Navigation: Navigate through axial, sagittal, and coronal slices.
• Session Statistics: View lesion counts, volumes, new lesions, and volume change metrics.
• Demo Data: Includes sample data for testing the application.

---

Technical Details

Lesion Matching Algorithm

The core analysis logic is implemented in utils/lesionAnalysis.js and consists of two main functions:

1. Connected Component Labeling (findConnectedComponents)

Segments a binary mask into individual lesion objects using iterative Union-Find.

Parameter	Value	Description
Threshold	> 0.8	Voxels above this probability are considered lesion
Connectivity	26-neighbor	Full 3D connectivity (including diagonals)
Minimum Size	≥ 3 voxels	Components smaller than this are discarded

Output: A labeled mask where each lesion has a unique integer ID, plus a list of lesion objects with centroid, volume, and ID.

2. Lesion Pair Analysis (analyzeLesionPair)

Matches lesions between two timepoints using graph-based transitive closure.

Algorithm Steps:

1. Label Components: Run findConnectedComponents on both masks independently.
2. Build Overlap Graph:
   • Nodes: Each component from TP1 ("1:ID") and TP2 ("2:ID").
   • Edges: If any voxel has both a TP1 label and a TP2 label, connect those nodes.
3. Union-Find Grouping: Use Union-Find to merge overlapping components into "Unified Lesions".
4. Classification: For each Unified Lesion:
   • New: No TP1 components, only TP2.
   • Gone: Only TP1 components, no TP2.
   • Growing: TP2 volume > TP1 volume.
   • Shrinking: TP2 volume < TP1 volume.
   • Static: TP1 volume = TP2 volume.

Unified Lesion Count: The number of graph nodes with at least one component at each timepoint. This handles merges (2 TP1 blobs → 1 TP2 blob) and splits correctly.

---

Session Statistics

Displayed in the UI via components/SessionStats.js:

Metric	Calculation
Lesion Count	Unified Lesion Count (not raw blob count)
Lesion Volume	Sum of all lesion voxels × voxel volume (mm³)
New Lesions	Count of Unified Lesions with status = 'new'
Vol Change	(TP2 Volume - TP1 Volume) / TP1 Volume × 100%

---

Project Structure

lesionViewOnline/
├── App.js                    # Main application entry point
├── components/
│   ├── DataLoadModal.js      # File upload and demo data loading
│   └── SessionStats.js       # Statistics panel UI
├── utils/
│   ├── lesionAnalysis.js     # Core lesion matching algorithm
│   ├── niftiUtils.js         # NIfTI file parsing utilities
│   └── colorUtils.js         # Color mapping for visualization
├── public/
│   └── demo/                 # Demo NIfTI data files
└── package.json

---

Dependencies

Package	Purpose
expo	React Native for Web framework
nifti-reader-js	Parse NIfTI files in browser
pako	Decompress gzipped NIfTI files

---

Usage

Development

npm install
npx expo start --web

Loading Data

1. Click Load Demo Data for sample brain data, or
2. Click Load Custom Data to select your own NIfTI files:
   • FLAIR TP1 (baseline)
   • FLAIR TP2 (follow-up)
   • Lesion Mask TP1
   • Lesion Mask TP2

Viewing Statistics

After loading data, the Session Stats panel shows:

• Lesion Count (TP1 → TP2)
• Lesion Volume (mm³)
• Number of New Lesions
• Volume Change (%)

Use the filter buttons (ALL, NEW, ENLARGING, STABLE) to highlight specific lesion types.

---

Algorithm Validation

The lesion matching algorithm has been validated against the Python implementation in laminateReport. Both implementations produce identical:

• Unified Lesion Counts
• New Lesion Detection
• Volume Change Statistics

The Python implementation (laminateReport/src/lesion_analysis.py) is a direct port of this JavaScript code.

---

License

MIT