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

Abstract

Abstract.A novel framework for robust 3D tracing in Electron Micrographs is presented. The proposed framework is built using ideas from hypergraph diffusion, and achieves two main objectives. Firstly, the approach scales to trace hundreds of targets without noticeable increase in runtime complexity. Secondly, the framework yields flexibility to fuse top down (global cues as hyperedges) and bottom up (local superpixels as nodes) information. Subsequently, a procedure for auto-seeding to initialize the tracing procedure is proposed. The paper concludes with experimental validation on a challenging large scale tracing problem for simultaneously tracing 95 structures, illustrating applicability of the proposed algorithm.

Robust Segmentation Based Tracing Using an Adaptive Wrapper for Inducing Priors

“…There has been a lot of interest in Electron Micrograph segmentation and tracing [5], [10], [23]- [26]. The primary motivation behind the proposed technique is the 3D tracing problem in retinal connectome data [27], a problem (data) not solved by any of the above referred techniques.…”

“…There exists substantial prior work in segmentation based tracing, with frameworks ranging from shortest path based [2], watersheds [3], [4], random walkers [5], active contours based snakes [6], geodesic active contours, vector flows [7], active contours without edges [8], and discrete valued Markov Random Fields (MRFs) inferred using graph cuts [9], [10]. The aim of this paper is to present a method that is capable of wrapping around a non-parametric segmentation technique, thereby achieving two objectives: Firstly, the proposed technique is capable of embedding high level priors into the tracing algorithm using free parameters of the base segmenter.…”

2D hidden Markov model with spatially adaptive state-space for tracing many cells in image sequence

“…Existing methods usually formulate the 3D reconstruction problem as one of grouping superpixels or supervoxels using various graphical models. Graph-based segmentation is utilized in [5] to obtain a 2D over-segmentation of every section, and a hypergraph framework is designed to efficiently solve the grouping problem. In [6], a probability map of membrane detection is learned by means of a random forest, and watershed segmentation is performed on top of the probability map to produce supervoxels that are then further grouped to reconstruct the 3D model.…”

Robust biological image sequence analysis using graph based approaches

“…The auxiliary algorithm used in our work employs boosted edge learning [10] / watersheds for generating homogeneous 2D image regions followed by shortest path computations on the volume, yielding higher order grouping constraints on superpixels. The reader is referred to [11], [12] for details pertaining to the cost function unifying top down and bottom up cues, and related inference issues.…”

“…Observe the severe clutter, object deformations and imaging artifacts that make the problem very challenging. We further benchmark the proposed approach with state of art tracing techniques, see [11] for more details. An important finding from our experiments is a marked rise in performance when hyperedges are utilized (F-Measure: 0.78), as opposed to using pairwise edges alone (F-Measure: 0.23).…”