Accurate Neuronal Soma Segmentation Using 3D Multi-Task Learning U-Shaped Fully Convolutional Neural Networks.

Tianyu Hu,Xiaofeng Xu,Shangbin Chen,Qian Liu

doi:10.3389/fnana.2020.592806

Abstract

Neuronal soma segmentation is a crucial step for the quantitative analysis of neuronal morphology. Automated neuronal soma segmentation methods have opened up the opportunity to improve the time-consuming manual labeling required during the neuronal soma morphology reconstruction for large-scale images. However, the presence of touching neuronal somata and variable soma shapes in images brings challenges for automated algorithms. This study proposes a neuronal soma segmentation method combining 3D U-shaped fully convolutional neural networks with multi-task learning. Compared to existing methods, this technique applies multi-task learning to predict the soma boundary to split touching somata, and adopts U-shaped architecture convolutional neural network which is effective for a limited dataset. The contour-aware multi-task learning framework is applied to the proposed method to predict the masks of neuronal somata and boundaries simultaneously. In addition, a spatial attention module is embedded into the multi-task model to improve neuronal soma segmentation results. The Nissl-stained dataset captured by the micro-optical sectioning tomography system is used to validate the proposed method. Following comparison to four existing segmentation models, the proposed method outperforms the others notably in both localization and segmentation. The novel method has potential for high-throughput neuronal soma segmentation in large-scale optical imaging data for neuron morphology quantitative analysis.

Highlights

Neuron morphology is crucial for brain function research, such as electrophysiology simulation, connectome, and neuron type classification (Svoboda, 2011)
The backbone of the proposed fully convolutional neural networks is designed based on U-shaped architecture which is effective for training in limited data
The pre- and post-processing steps of this study are applied in other 3D convolutional neural networks (CNNs), and the weighted cross-entropy loss function (Ronneberger et al, 2015) proposed for cell segmentation is used in other 3D CNNs

Summary

Introduction

Neuron morphology is crucial for brain function research, such as electrophysiology simulation, connectome, and neuron type classification (Svoboda, 2011). Accurate Neuronal Soma Segmentation (Gong et al, 2013; Wu et al, 2014), it is possible to acquire high-resolution large-scale neuron imaging datasets. As an alternative method, automated neuronal soma segmentation is highly efficient, and provides accurate results for neuronal somata morphology reconstruction (Meijering, 2010). The complexity of the optical imaging datasets causes many challenges for automated neuronal soma segmentation algorithms. The images consistently show touching neuronal somata that are clustered in several local regions, with unclear boundaries between them. For these reasons, it is hard to localize or divide these neuronal into individual soma. The intensity threshold method has difficulty in segmenting touching cells. While an improved threshold method using multi-level intensity to separate touching cells has been previously proposed (Keenan et al, 2000), since the touching cells have similar brightness and adjacent position (He et al, 2014), it remains challenging to perform accurate segmentation in this manner

Methods

Results

Discussion

Conclusion