Segmentation fusion for connectomics

Abstract

We address the problem of automatic 3D segmentation of a stack of electron microscopy sections of brain tissue. Unlike previous efforts, where the reconstruction is usually done on a section-to-section basis, or by the agglomerative clustering of 2D segments, we leverage information from the entire volume to obtain a globally optimal 3D segmentation. To do this, we formulate the segmentation as the solution to a fusion problem. We first enumerate multiple possible 2D segmentations for each section in the stack, and a set of 3D links that may connect segments across consecutive sections. We then identify the fusion of segments and links that provide the most globally consistent segmentation of the stack. We show that this two-step approach of pre-enumeration and posterior fusion yields significant advantages and provides state-of-the-art reconstruction results. Finally, as part of this method, we also introduce a robust rotationally-invariant set of features that we use to learn and enumerate the above 2D segmentations. Our features outperform previous connectomic-specific descriptors without relying on a large set of heuristics or manually designed filter banks.

Optimization-Based Artifact Correction for Electron Microscopy Image Stacks

“…Imaging artifacts, most notably discontinuities along the section (z) axis, have so far limited the use of image volumes acquired by ATUM-based SEM, one of the most promising high-throughput volume electron microscopy techniques, to essentially 2.5-D analysis [34,10]. Our limited assumptions about the data and distortion process lead naturally to a simple but highly effective optimization-based procedure: our method EMISAC appears to eliminate all visible discontinuities, without any loss of intra-section detail.…”

Scalable Tracing of Electron Micrographs by Fusing Top Down and Bottom Up Cues Using Hypergraph Diffusion

“…The data used in this work is very different and is from a rabbit retina (noisy regional texture based). Techniques similar to [12] did not perform well on our datasets, leading us to compare with the state of the art on rabbit retina and relevant tracing techniques. Our work is intended as a scalable replacement to the graph cut solvers used in [6], as will be established by experimental results.…”

“…The works by [7,8,6,12] are good sources of reference for EM image analysis. Further, [5] utilized hypergraphs for unsupervised video segmentation, in contrast to the supervised case the proposed approach deals with.…”

“…Our claim of originality is in the framework comprising the salient aspects listed above. In related work, [12] propose an interesting technique utilizing pairwise segment interactions on EM data from a mouse (gradient based), but do not lay emphasis on user interaction or scalability. The data used in this work is very different and is from a rabbit retina (noisy regional texture based).…”

Anisotropic ssTEM Image Segmentation Using Dense Correspondence across Sections

“…The third approach [15] generates many, possibly contradictory, segmentation hypotheses in individual sections and combine them in order to optimize the global agreement functional defined on the whole stack. In contrast to this approach, we are not dealing with given segmentation hypotheses, but incorporate the context from neighboring sections to improve the segmentation of every single section.…”