MorphGrower: A Synchronized Layer-by-layer Growing Approach for Plausible Neuronal Morphology Generation

TL;DR

MorphGrower addresses the challenge of generating plausible neuronal morphologies by adopting a synchronized layer-by-layer growth framework that emits branch pairs conditioned on previously generated structure. It combines a CVAE with vMF latent sampling, a Tree-structure aware GNN for global conditioning, and EMA-based local conditioning to capture growth dependencies, enforcing topological validity (soma as the sole multi-branch node). Empirical results on four mouse datasets show superior morphology statistics over MorphVAE, improved generation plausibility via a Real/Fake classifier, and electrophysiological simulations demonstrating neuroscience-level plausibility. The approach yields growth-stage snapshots and offers a scalable data-generation pathway to augment neuronal morphology datasets for large-scale brain simulations, albeit with limitations related to coordinate-only data and simplified dynamics. Overall, MorphGrower represents a substantial step toward realistic, growth-aware neuron morphology synthesis with potential for population-level modeling and broader bioimaging applications.

Abstract

Neuronal morphology is essential for studying brain functioning and understanding neurodegenerative disorders. As acquiring real-world morphology data is expensive, computational approaches for morphology generation have been studied. Traditional methods heavily rely on expert-set rules and parameter tuning, making it difficult to generalize across different types of morphologies. Recently, MorphVAE was introduced as the sole learning-based method, but its generated morphologies lack plausibility, i.e., they do not appear realistic enough and most of the generated samples are topologically invalid. To fill this gap, this paper proposes MorphGrower, which mimicks the neuron natural growth mechanism for generation. Specifically, MorphGrower generates morphologies layer by layer, with each subsequent layer conditioned on the previously generated structure. During each layer generation, MorphGrower utilizes a pair of sibling branches as the basic generation block and generates branch pairs synchronously. This approach ensures topological validity and allows for fine-grained generation, thereby enhancing the realism of the final generated morphologies. Results on four real-world datasets demonstrate that MorphGrower outperforms MorphVAE by a notable margin. Importantly, the electrophysiological response simulation demonstrates the plausibility of our generated samples from a neuroscience perspective. Our code is available at https://github.com/Thinklab-SJTU/MorphGrower.

Figures (31)

• Figure 1: Morphology in node and branch views. The node representing the soma, the nodes signifying bifurcations or tips, and the regular nodes along each branch but not at their ends are highlighted in different colors for distinction.
• Figure 2: Overview of MorphGrower. It takes the branch pair and its previous layers as inputs. The branch pairs as well as the previous layers as conditions determine the mean direction $\mu$ of latent space which follows a von-Mises Fisher distribution with fixed variance $\kappa$. Latent variables are then sampled from the distribution $\mathbf{z}_i\sim \mathrm{vMF}(\mu, \kappa)$. Finally, the decoder reconstructs the branch pairs from the latent variable $\mathbf{z}$ and the given condition. In inference, the model is called regressively, taking the generated subtree $\hat{T}^{(i)}$ and the $(i+1)$-th layer $L_{i+1}$ of the reference morphology as input and outputting a new layer $\hat{L}_{i+1}$ of the final generated morphology.
• Figure 3: Distributions of four morphological metrics: MPD, BPL, MED and CTT on VPM dataset.
• Figure 4: Comparison of averaged electrophysiological recordings on real and generated morphology samples. The term MP on the vertical axis stands for Membrane Potential.
• Figure 5: Projections onto $xy$ plane of three adjacent intermediate morphologies of a generated sample from RGC. For each $\hat{T}^{(j)}$, the newly generated layer $\hat{L}_j$ and the intermediate morphology generated after the last step $\hat{T}^{(j-1)}$ are highlighted in pink and blue, respectively.

Theorems & Definitions (2)

• Definition 2.1: Soma & Tip & Bifurcation
• Definition 2.2: Compartment & Branch