Synaptic Partner Prediction from Point Annotations in Insect Brains

Abstract

High-throughput electron microscopy allows recording of large stacks of neural tissue with sufficient resolution to extract the wiring diagram of the underlying neural network. Current efforts to automate this process focus mainly on the segmentation of neurons. However, in order to recover a wiring diagram, synaptic partners need to be identified as well. This is especially challenging in insect brains like Drosophila melanogaster, where one presynaptic site is associated with multiple postsynaptic elements. Here we propose a 3D U-Net architecture to directly identify pairs of voxels that are pre-and postsynaptic to each other. To that end, we formulate the problem of synaptic partner identification as a classification problem on long-range edges between voxels to encode both the presence of a synaptic pair and its direction. This formulation allows us to directly learn from synaptic point annotations instead of more expensive voxel-based synaptic cleft or vesicle annotations. We evaluate our method on the MICCAI 2016 CREMI challenge and improve over the current state of the art, producing 3% fewer errors than the next best method.

Automated Reconstruction of a Serial-Section EM Drosophila Brain with Flood-Filling Networks and Local Realignment

“…Moreover, some highly useful biological inferences, such as identification of neuronal cell type, can be performed using FAFB-FFN1 in minutes or seconds per neuron, as compared to hours in the purely manual case. Experiments also demonstrated the applicability of FAFB-FFN1 to circuit tracing and connectomic analyses; these studies combine segmentation with an accounting of the synaptic connections between neurons, for which large-scale automated methods on Drosophila tissue are an active research area (Kreshuk et al 2015;Buhmann et al 2018;Heinrich et al 2018;Huang, Scheffer, and Plaza 2018) .…”

“…6), but its upstream and downstream partner neurons had not. We further assumed that all input and output synapse locations had been marked; in this case marking was done manually, but an automated approach (Kreshuk et al 2015;Buhmann et al 2018;Heinrich et al 2018;Huang, Scheffer, and Plaza 2018) would also be viable.…”

Local Shape Descriptors for Neuron Segmentation

“…Consequently, automatic methods for the reconstruction of neurons and identification of synapses have been developed. Over the past decade, methods targeting relatively small volumes have pioneered the reconstruction of neurons (Turaga et al, 2010;Lee et al, 2017), and synapses (Kreshuk et al, 2015;Buhmann et al, 2018). More recently, these efforts have been improved to tackle the challenges of large datasets for neurons (Januszewski et al, 2018;Funke et al, 2019;Dorkenwald et al, 2019;Li et al, 2019), synaptic clefts (Heinrich et al, 2018), and synaptic partners (Huang et al, 2018;Buhmann et al, 2020).…”

Automatic Detection of Synaptic Partners in a Whole-BrainDrosophilaEM Dataset

“…Huang et al (2018) solve a similar classification task with a multilayer perceptron and local context features. In contrast to these two-step approaches, Buhmann et al (2018) jointly detect pre-and post-synaptic sites and infer their connectivity via longrange affinity edges using a single CNN architecture.…”

“…More specifically, we use a 3D U-Net CNN (Ronneberger et al, 2015) to predict for each voxel in a volume whether it is part of a post-synaptic site and, if it is, an offset vector pointing from this voxel to the corresponding pre-synaptic site. The method we introduce here is similar to our previous work (Buhmann et al, 2018) in that it does not require explicit annotation or detection of synapse features (such as T-bars or clefts), but instead predicts synaptic partners from raw EM data directly. This design choice has two practically important implications: First, our method can be trained from pre-and post-synaptic point annotations alone, which reduces the effort needed to annotate future training datasets.…”