Overview
The Discrete Wavelet Transform (DWT) is a foundational mathematical matrix decomposition technique widely applied across medical image processing, neural signal analysis, satellite remote sensing, and lossy-to-lossless image compression standards such as JPEG2000. Unlike classical Fourier transformations—which express signals purely as a summation of infinite sinusoidal oscillations spanning the global time domain—wavelet transforms decompose complex two-dimensional spatial intensity distributions into localized multi-resolution frequency components using compactly supported basis functions.
In diagnostic neuroimaging, high-resolution Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) scans capture minute anatomical structures, subtle tissue contrast boundaries, ischemic infarctions, and intracranial hematomas. Efficient storage and rapid transmission of these volumetric datasets across clinical Picture Archiving and Communication Systems (PACS) require lossy compression schemes that preserve critical diagnostic features while aggressively pruning perceptually redundant spatial frequencies.
How to Use
The application features a visual split workspace displaying four synchronous diagnostic canvases alongside an interactive control console:
- Clinical Image File Dataset: Choose directly from authentic neuroimaging scans in the dataset, including multi-slice contrast-enhanced Brain MRIs, acute ischemic strokes, intracerebral hemorrhages, lobar hemorrhages, pontine brainstem infarctions, and thalamic lacunar lesions courtesy of Radiopaedia.org. You can also upload custom DICOM-converted files directly into browser memory.
- Wavelet Basis Family: Choose the mathematical wavelet basis used for row-column filtering (Haar/db1, Daubechies 2, Symlets 2, or Biorthogonal 1.3).
- Decomposition Levels Slider: Adjust the recursive pyramidal depth (1 to 5 levels). Each level recursively decomposes the low-frequency approximation subband ($LL$) into four smaller subquadrants ($LL, LH, HL, HH$).
- Quantization Budget (Bits) Slider: Manipulate the bit-depth allocated to represent wavelet coefficients after forward transformation. Lower bit depths introduce uniform scalar quantization coarse steps and zero out small detail coefficients.
- Start Demo & Reset Baseline: Click "Start Demo" to run an automated tour through different clinical images, wavelet bases, and quantization levels. Click "Reset Baseline" to instantly restore all controls to their default state.
Technical Details
The 2D Discrete Wavelet Transform operates via separable 1D filtering across image rows and columns using a pair of quadrature mirror filters: a low-pass scaling filter $h[n]$ and a high-pass wavelet filter $g[n]$. For a 2D intensity grid $f(x,y)$, a single-level decomposition computes four distinct subband quadrants:
1. LL (Approximation): Low-pass filtered along both rows and columns ($h * h$), capturing the smooth coarse structural topology.
2. LH (Horizontal Details): Low-pass along rows, high-pass along columns ($h * g$), isolating horizontal edges and boundaries.
3. HL (Vertical Details): High-pass along rows, low-pass along columns ($g * h$), isolating vertical anatomical borders.
4. HH (Diagonal Details): High-pass along both axes ($g * g$), capturing high-frequency diagonal noise and point singularities.
Future Directions
Upcoming iterations of this interactive module aim to integrate 3D directional wavelets (such as Dual-Tree Complex Wavelets, Contourlets, and Curvelets) to improve anisotropic contour representations in volumetric CT and MRI multi-slice datasets.
Related Interactive Environments
Images courtesy of Radiopaedia.org.