Efficient and Effective 3D Visualization of Internal Neuroimage Structures with NiiVue

A novel normalized logarithmic gradient magnitude approach, implemented in NiiVue, enhances visualization of internal structures while maintaining real-time performance across devices.

Introduction

• Neuroimaging Visualization Challenge:
  • Visualizing internal brain structures is critical for understanding complex neuroanatomy, functional relationships, and pathology detection.
  • Our goal is to provide advanced visualization techniques through an intuitive web interface accessible to researchers without specialized hardware or software installations.
• NiiVue Capabilities:
  • NiiVue [1] is a web-based visualization tool for neuroimaging that runs on any operating system and web device (phone, tablet, computer).
  • Uniquely displays multiple neuroimaging datatypes simultaneously: voxels, meshes, tractography streamlines, statistical maps, and connectomes.
• Gradient Opacity Role: Traditional volume rendering often obscures internal structures behind high-intensity regions; gradient opacity allows visualization through tissue based on structural boundaries. [2]

Methods

• Normalized Logarithmic Gradient Approach:
  • Applies logarithmic transformation to compress wide dynamic range of gradient magnitudes
  • Normalizes values to range for intuitive user control
    $$o = \left( \log_2\left(G_x^2 + G_y^2 + G_z^2 + \frac{1}{255^2 \cdot 8}\right) + \log_2\left(\frac{1}{255^2 \cdot 8}\right) \right)^{8.0}$$
    Where final exponentiation to 8.0 adds translucency to rendering.
• Gradient Magnitude Calculation:
  • First-order gradients computed using 3D Sobel-approximation operators
  • Optionally uses second-order gradients based on Taylor Series expansion for enhanced detail [3]: \(f^1_0 = \frac{f_{-2} - 8 f_{-1} + 8 f_1 - f_2}{ 12 h }\)



Click to interact

• Real-Time, GPGPU Performance [4]:
  • Pre-computes gradients using WebGL2 fragment shaders for parallel processing
  • Leverages GPU texture interpolation hardware for optimized sampling
  • Encodes 3D gradient orientation and magnitude in only 4 bytes total per voxel
  • One-time preprocessing cost of ~100ms for typical brain volumes
  • Minimal impact on frame rate during interactive viewing