Automated 3D Neuron Tracing with Precise Branch Erasing and Confidence Controlled Back-Tracking

Abstract

The automatic reconstruction of single neuron cells from microscopic images is essential to enabling large-scale datadriven investigations in neuron morphology research. However, few previous methods were able to generate satisfactory results automatically from 3D microscopic images without human intervention. In this study, we developed a new algorithm for automatic 3D neuron reconstruction. The main idea of the proposed algorithm is to iteratively track backwards from the potential neuronal termini to the soma centre. An online confidence score is computed to decide if a tracing iteration should be stopped and discarded from the final reconstruction. The performance improvements comparing to the previous methods are mainly introduced by a more accurate estimation of the traced area and the confidence controlled back-tracking algorithm. The proposed algorithm supports large-scale batch-processing by requiring only one hyper-parameter for background segmentation. We benchtested the proposed algorithm on the images obtained from both the DIADEM challenge and the BigNeuron challenge. Our proposed algorithm achieved the state-of-the-art results.

Optimization of Traced Neuron Skeleton Using Lasso-Based Model

“…In neuron reconstruction, many methods focus on tracing neurite segments in challenging cases, i.e., fuzzy or broken segments in noisy environments. Some best available methods are listed here, such as graph-based models (Peng et al, 2011; Turetken et al, 2011; Basu et al, 2013; De et al, 2016), principle curve models (Bas and Erdogmus, 2011; Li et al, 2016; Quan et al, 2016), iterative back-tracking (Liu et al, 2018), optimization models (Zhao et al, 2011; Skibbe et al, 2018), minimal path approaches (Lee et al, 2012; Yang et al, 2018), learning structured features (Gu et al, 2017) etc. These methods can automatically extract skeleton points and branching points, which drop into three-dimensional tubular region in neuronal images and construct the geometrical structure of the neuron.…”

Exploring highly reliable substructures in auto-reconstructions of a neuron

“…To quantitatively evaluate the improvement of individual motifs’ accuracy, we first define a node in a SWC file (individual motifs or automatic reconstructions) as an accurate node if its distance to the nearest ground truth node is no more than 4 voxels [ 14 ], and then the precision rate of a reconstruction or its individual motifs is defined as the ratio of the number of accurate nodes to its total node number. For reconstructions from each species generated by each of above four automatic tracing methods, Table 2 gives their average precision rate and that of their corresponding individual motifs.…”

“…In recent years, many automatic methods and tools have been developed for digital reconstruction of neurons, such as automatic contour extraction [ 2 ], APP1 [ 3 , 4 ], APP2 [ 5 ], MOST [ 6 ], SmartTracing [ 7 ], Ray casting [ 8 ], tTuFF [ 9 ], Rivulet [ 10 ], SparseTracer [ 11 ], M-AMST [ 12 ], Ensemble Neuron Tracer [ 13 ], Rivulet2 [ 14 ], FMST [ 15 ], MOST-based repairer [ 16 ], 3-D upgraded ray-shooting [ 17 ], and so on. Two memorabilia were held to promote the research of automatic tracing technologies.…”

Review of advances and prospects in neuron reconstruction

“…Ultra Tracer 软 件 [75] , 将 大 尺 度 数 据 分 析 方 法 [76] (Terafly)与多种传统单神经元重建算法结合, 可在较 大尺度内(2111×3403×291 voxels)快速重建单神经元 的树突形态; 华盛顿州立大学和艾伦脑科学研究所 和艾伦研究所的科研人员基于三维卷积神经网络开 发的神经元重建方法, 能有效地重建缺断、噪声较大 区域内的神经纤维, 该方法对于重建这类挑战性图 像 效 果 很 好 [77] ; 京 都 大 学 发 展 了 probabilistic axon tracking(PAT)算法 [78] , 它基于蒙特卡洛优化模型追 踪TB量级下交叉、 缠绕和扭曲的轴突纤维, 改善了现 有重建方法对于密集轴突追踪不准确的现状; 悉尼 大学开发的Rivulet系列算法 [79,80] 基于多模式快速行…”