Density-encoded line integral convolution: polarisation optical axis tractography using centroidal Voronoi tessellation

Darven Murali Tharan, Marco Bonesi, Daniel Everett, Cushla McGoverin, Sue McGlashan, Ashvin Thambyah, Frédérique Vanholsbeeck · 2025

Read Density-encoded line integral convolution: polarisation optical axis tractography using centroidal Voronoi tessellation by Darven Murali Tharan, Marco Bonesi, Daniel Everett, Cushla McGoverin, Sue McGlashan, Ashvin Thambyah, Frédérique Vanholsbeeck (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-02060.txt.

Introduction

Tharan and six co-authors present DELIC, a 2025 method for drawing complex optical axis fields from polarimetry data. It extends the classic line integral convolution technique with centroidal Voronoi tessellation, a way of dividing space into balanced cells. The density encoding makes crowded and tangled axis patterns clearer and easier to read. The authors aim the tool at researchers who need to see fiber and axis structure in optical data.

Catalog record

CREATORDarven Murali Tharan, Marco Bonesi, Daniel Everett, Cushla McGoverin, Sue McGlashan, Ashvin Thambyah, Frédérique Vanholsbeeck
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-02060

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Density-encoded line integral convolution: polarisation optical axis tractography using centroidal Voronoi tessellation Darven Murali Tharan, Marco Bonesi, Daniel Everett, Cushla McGoverin, Sue McGlashan, Ashvin Thambyah, Frédérique Vanholsbeeck Abstract Visualising complex polarimetry optical axis fields is challenging. We introduce density-encoded line integral convolution (DELIC), a novel approach that builds on the classic line integral convolution algorithm by incorporating the principles of centroidal Voronoi tessellation, enabling clearer and more interpretable representations of complex optical axis fields. I Introduction Vector field visualisation plays a critical role in scientific research, aiding in the analysis of complex flow patterns. Such tools are widely used in biomedical imaging, particularly in tractography, where the structural pathways of neuronal axons, muscle fibres, and collagen are mapped. Biomedical polarimetry, the study of how tissue alters the polarisation state of light [ 1 ] , has become an essential tool for tractography in birefringent tissues [ 2 , 3 , 4 ] . Many biological tissues, like those composed of collagen or axons, consist of highly organised fibrous structures that exhibit birefringence, altering the polarisation state of light. Polarimetry enables measurement of two key properties: birefringence strength, which reflects fibrous density and organisation, and the optical axis, indicating the orientation of these structures. In polarimetry-based tractography, methods such as streamlines and glyphs are commonly used to visualise fibrous orientation. These methods can clutter visualisations in dense or complex fields, making interpretation challenging [ 5 ] . However, these methods remain popular due to their simplicity and widespread availability, such as MATLAB’s stream2 , Python’s streamplot , and R’s geom_path . Line integral convolution (LIC) provides a more sophisticated approach by convolving a noise texture along streamlines, offering a clutter-free and accurate representation of the fine details in complex vector fields [ 6 ] . Despite its advantages, LIC has seen limited use in polarimetry-based tractography. One reason for this could be the high density and high-frequency content of LIC images, which may overwhelm and become difficult to interpret [ 7 ] . In this letter, we introduce density-encoded line integral convolution (DELIC), a novel algorithm specifically designed for visualising birefringent tissue inspired by the traditional LIC technique. Unlike LIC, which uses a noise texture that spans the entire domain, DELIC employs a sparse noise texture. This sparsity helps address the issue of LIC’s dense and high frequency content, which can make interpretation difficult [ 7 ] . While reducing texture density may seem like a loss of detail compared to LIC, DELIC compensates by selectively concentrating texture density in regions of interest, such as areas of high birefringence, where the tissue structural information is assumed most significant. By modulating texture density based on birefringence strength, DELIC enhances the clarity of critical regions while simplifying less important areas, reducing visual clutter. This targeted approach preserves essential structural details, offering clearer and more interpretable visualisations with minimal loss of information. We demonstrate the effectiveness of DELIC by direct comparison with LIC using polarimetry data acquired through a custom-built polarisation-sensitive optical coherence tomography (PS-OCT) system. Optical Coherence Tomography (OCT) is non-invasive imaging technique that is the optical equivalent to ultrasonography, capable of generating cross-sectional (sagittal plane) images of biological tissue [ 8 ] . PS-OCT is a functional extension of this technique that allows measurement of tissue birefringence and the optical axis of birefringence [ 9 , 10 ] . Our results highlight how DELIC enhances the visualisation of fibrillar architecture, offering improved clarity compared to LIC. This approach represents a significant advancement in polarimetry visualisation, offering a powerful and effective tool for the clear and accurate interpretation of polarimetry data. DELIC has the potential to be applied both in scientific research and clinical settings, enabling improved in vivo analysis and exploration of birefringent tissues. II Methods II-A Line Integral Convolution (LIC) Consider a vector field \mathbf{V}(\mathbf{r}) over a spatial domain \Omega\subset\mathbb{R}^{d} , where \mathbf{r} represents a position. Streamlines are computed by integrating the vector field from each point \mathbf{r} , producing paths that follow the direction of the field. For each point, the path is parametrised as \mathbf{s}(t;\mathbf{r}) , where t is a parameter along the path: \mathbf{s}(t;\mathbf{r})=\mathbf{r}+\text{sgn}(t)\int_{0}^{|t|}\mathbf{V}(\mathbf{s}(\tau;\mathbf{r}))\,d\tau, (1) where \text{sgn}(t) is the signum function. In this work, we use the 4th-order Runge-Kutta (RK4) method for numerical integration. A randomly generated noise texture, I_{n}(\mathbf{r}) , is then convolved along the streamline using a kernel k(t) , which in this work is a Gaussian window. The LIC value at a point \mathbf{r} , denoted as I_{LIC}(\mathbf{r}) , is computed as: I_{LIC}(\mathbf{r})=\frac{1}{K}\int_{-T}^{T}I_{n}(\mathbf{s}(t;\mathbf{r}))k(t)\,dt, (2) where K=\int_{-T}^{T}k(t)\,dt and T is the convolution period. II-B Density encoded line integral convolution While traditional LIC initialises with a random noise texture that fills the entire domain, \Omega , DELIC uses a sparse texture field, where values are assigned only to a select set of seed points scattered across the domain. The rest of the domain remains empty or zero-valued. This reduces clutter and improves clarity in the resultant tractogram, making the vector field direction easier to interpret [ 7 ] . By modulating seed point density based on an underlying scalar field, DELIC emphasises regions of interest such as areas of high birefringence strength , ensuring that key features are highlighted without losing detail. Moving forward, unless stated otherwise, we denote the birefringence scalar field used to generate a DELIC tractogram as the encoding field, \rho(\mathbf{x}) . Below, we start by giving a brief summary of the entire DELIC processing pipeline which leads to an in-depth description in the subsequent sections. • Extract and Process Data: Retrieve birefringence strength (Fig. 1 A) and optical axis orientation data (Fig. 3 A) and define a vector field of the optical axis. • Seed Point Generation: Normalize birefringence to define a probability density function (PDF), then apply rejection sampling to generate seed points reflecting the PDF (Fig. 1 B), with a minimum distance of d_{\mbox{sep}} between each seed. • Seed Point Refinement: Refine the initial seed points to better reflect the PDF. (Fig. 1 C). • Texture generation: Assign random grey values to the refined seed points using Sobol quasi-random generation [ 11 ] to create a texture image (Fig. 1 D). • LIC Generation: Perform LIC on this texture (Fig. 3 D). Optionally, apply unsharp masking to enhance sharpness and histogram stretching to improve contrast. Colour mapping based on the optical axis orientation may also be applied. II-B 1 DELIC initialisation Strategic placement of seed points is required for DELIC to ensure accurate representation of the encoding field. We begin with an initial point distribution generated via basic rejection sampling of the encoding field, approximating the desired density function (Fig. 1 B). We enforce a minimum distance of d_{\mbox{sep}} between each seed. This is the primary method for controlling the global density of seeds for the algorithm. This initial distribution of points, denoted \{\mathbf{r}_{i}^{(0)}\}_{i=1}^{n} , serves as the starting point for further refinement. Next, we utilise the principles of centroidal Voronoi tessellation (CVT) [ 12 ] to refine \{\mathbf{r}_{i}^{(0)}\}_{i=1}^{n} so that it more accurately describes the encoding field (Fig. 1 C). Fig. 1: (A) Normalised birefringence strength image from skin. Vertical and horizontal scale bars are 500 \mu m long. (B) Initial seed point distribution for DELIC, generated from (A). The

Read the complete text