Online Multi-spectral Neuron Tracing

TL;DR

This paper tackles neuron tracing in densely labeled multi-spectral 3D stacks by proposing xBTracer, an online, training-free tracer that starts from a bounding box and continuously updates a global color model and an enhanced discriminative correlation filter $\mathbf{w}$. The method introduces three designs: an Alpha Channel Enhanced DCF for compact reconstructions, a Cross-section Determination Module that selects stable cross-planes using a memory bank, and a Bifurcation Determination Module that uses an A* path-based cost to detect branching. Empirical results on simulated and real multi-spectral data show significant gains over baselines in metrics such as Variation of Segment Alignment (VSA), Mean Fitting Distance (MFD), Bifurcation Retrieval Rate (BRR), and DIADEM, with demonstrated generalization to other modalities like $fMOST$. The approach reduces user configuration, eliminates annotation requirements, and supports fast, accurate neuron reconstructions across imaging modalities, marking a practical advance for automated neuron tracing in complex spectral datasets.

Abstract

In this paper, we propose an online multi-spectral neuron tracing method with uniquely designed modules, where no offline training are required. Our method is trained online to update our enhanced discriminative correlation filter to conglutinate the tracing process. This distinctive offline-training-free schema differentiates us from other training-dependent tracing approaches like deep learning methods since no annotation is needed for our method. Besides, compared to other tracing methods requiring complicated set-up such as for clustering and graph multi-cut, our approach is much easier to be applied to new images. In fact, it only needs a starting bounding box of the tracing neuron, significantly reducing users' configuration effort. Our extensive experiments show that our training-free and easy-configured methodology allows fast and accurate neuron reconstructions in multi-spectral images.

Figures (6)

• Figure 1: Overview of the components of our proposed online multi-spectral neuron tracing. Details of each component can be found in the Methodology section.
• Figure 2: Our alpha channel enables more compact reconstruction. The overlayed patches are obtained by applying the corresponding mask to the same image patch.
• Figure 3: An example of how to filter out objects for bifurcation. Here, the one object with white path is retained while the one with red path is abandoned.
• Figure 4: Diagrammatic sketch of modeling bifurcation. (Left) The overall trajectory (red line) in 3D. (Right) Detailed 2D movements.
• Figure 5: Comparisons to the most performing neuron tracing methods. We can observe more compact reconstruction using our method. The average area of the traced bboxes is $\sim$80 pixels, where we pad our traced patches to the same size for better visualization on the bottom right.