TraceMontage : a Method for Merging Multiple Independent Neuronal Traces

Aslan S. Dizaji, Logan A. Walker, Dawen CaibioRxiv 2019 · DOI

TL;DR

Mapping neuronal circuits is crucial for understanding brain function, but tracing individual neurons in dense, high-resolution images is challenging. TraceMontage is a new ImageJ/Fiji plugin that automatically merges multiple neuron traces from the same or adjacent image tiles, creating a unified and more accurate representation of neuronal structures. This tool is essential for scaling up connectomics research to larger and more complex neural systems.

What Problem Does This Paper Address?

The ultimate goal of connectomics is to map the intricate wiring of neural networks. Modern imaging techniques, like Brainbow labeling and advanced microscopy, allow us to capture incredibly detailed images of neural tissue. However, tracing the complex, branching structures of neurons within these dense images is a bottleneck. Often, to manage the sheer volume of data or for quality control, images are split into smaller tiles, and neurons are traced independently. This results in multiple, potentially overlapping, traces that need to be combined into a single, coherent map. Manually merging these traces is time-consuming and prone to errors. Existing tools lack a dedicated solution for this crucial post-processing step, especially when dealing with dense labeling where multiple neurons are intertwined.

Methods

TraceMontage is an ImageJ/Fiji plugin designed to automate the merging of neuronal traces. Its core functionality relies on graph theory. Each neuron trace is represented as a tree graph, where nodes represent points along the neuron's path and edges represent the connections. The plugin identifies overlapping branches between different traces by comparing their 3D spatial coordinates. It then uses a fuzzy-logic inspired algorithm to fuse these overlapping branches, preserving the overall tree structure of the neuron. If there are disagreements between traces (e.g., different interpretations of a branch), the plugin defaults to the trace designated as "primary" (the first input trace). Beyond merging, TraceMontage also provides consistency analysis metrics to evaluate the agreement between different tracing efforts.

Datasets

The study utilized microscope images from the hippocampus and primary visual cortex (V1) of Brainbow-labeled mice. These images were split into overlapping tiles to simulate a typical experimental setup where wide-field images are acquired in sections. The individual tiles were then traced using the nTracer software by trained biologists. These traced datasets, along with the original Brainbow images, were used to demonstrate and evaluate TraceMontage.

Key Insights and Ideas

  • The power of graph theory for merging: Representing neuronal traces as graphs allows for a robust and systematic way to identify and reconcile overlapping structures, moving beyond simple spatial overlap.
  • Collaborative tracing boosts accuracy and speed: By combining the efforts of multiple tracers (or multiple tracing passes), TraceMontage can improve the accuracy and efficiency of reconstructing densely labeled neural samples. This is akin to having multiple experts review and integrate their findings.
  • Bridging the gap in connectomics workflows: TraceMontage fills a critical gap in the post-processing pipeline for connectomics data, enabling researchers to scale their analyses to larger datasets.

Key Results

The paper demonstrates TraceMontage's ability to effectively merge horizontally and vertically aligned image tiles, as well as to reconcile traces from multiple independent tracing efforts on the same image. Figure 2 showcases an example where two semi-manually traced tiles of Parvalbumin-expressing basket cells were merged, clearly showing how the overlapping regions connect seamlessly. While specific quantitative metrics like percentage of merged traces or accuracy improvements are not detailed with concrete numbers in the main text, the qualitative results and the visual examples strongly suggest successful merging and improved continuity of neuronal structures. The authors highlight that the default parameter settings were used, implying that further optimization could potentially enhance the merging process.

Limitations and Future Work

The authors acknowledge several limitations:

  • Binary tree structure: TraceMontage is currently limited to structures that can be represented as a full binary tree, excluding non-binary structures like vasculature.
  • Trace exclusion: In cases where neurites are not traced in both input reconstructions, some original tracing information might be lost to maintain consistency. This is because the merging relies on branch endpoints, not internodes.
  • Arbitrary primary tracer: The current version arbitrarily designates the first input trace as "primary" in case of disagreements. Future versions could offer options to calculate an average or use more sophisticated methods for resolving conflicts.
  • Limited to two traces: The current algorithms are designed for merging two tracing results. Developing algorithms for merging more than two traces would be beneficial for creating highly accurate "gold-standard" datasets in collaborative settings.

Future work could address these limitations by incorporating more advanced graph matching algorithms, developing methods for handling non-binary structures, and implementing more sophisticated conflict resolution strategies.

Summary updated

Abstract

Abstract Background The ability to reconstruct neuronal networks, local microcircuits, or the entire connectome is a central goal of modern neuroscience. Recently, advancements in sample preparation (e.g., sample expansion and Brainbow labeling) and optical (e.g., confocal and light sheet) techniques have enabled the imaging of increasingly large neural systems with high contrast. Tracing neuronal structures from these images proves challenging, however, necessitating tools that integrate multiple neuronal traces, potentially derived by various methods, into one combined ( montaged ) result. New Method Here, we present TraceMontage, an ImageJ/Fiji plugin for the combination of multiple neuron traces of a single image, either redundantly or non-redundantly. Internally, it uses graph theory to connect topological patterns in the 3-D spatial coordinates of neuronal trees. The software generates a single output tracing file containing the montage traces of the input tracing files and provides several measures of consistency analysis among multiple tracers. Results and Comparison to existing method(s) To our knowledge, our software is the first dedicated method for the combination of tracing results. Combining multiple tracers increases the accuracy and speed of tracing of densely-labeled samples by harnessing collaborative effort. This utility is demonstrated using fluorescence microscope images from the hippocampus and primary visual cortex (V1) in Brainbow-labeled mice. Conclusions TraceMontage provides researchers the ability to combine neuronal tracing data generated by either the same or different method(s). As datasets become larger, the ability to trace images in this parallel manner will help connectomics scale to increasingly larger neural systems.
BibTeX
@misc{dizaji2019i,
  author = {Aslan S. Dizaji and Logan A. Walker and Dawen Cai},
  title = {{<i>TraceMontage</i> : a Method for Merging Multiple Independent Neuronal Traces}},
  year = {2019},
  doi = {10.1101/703900},
  howpublished = {bioRxiv},
}
Figure