Active Learning and Proofreading for Delineation of Curvilinear Structures

Abstract

Many state-of-the-art delineation methods rely on supervised machine learning algorithms. As a result, they require manually annotated training data, which is tedious to obtain. Furthermore, even minor classification errors may significantly affect the topology of the final result. In this paper we propose a generic approach to addressing both of these problems by taking into account the influence of a potential misclassification on the resulting delineation. In an Active Learning context, we identify parts of linear structures that should be annotated first in order to train a classifier effectively. In a proofreading context, we similarly find regions of the resulting reconstruction that should be verified in priority to obtain a nearly-perfect result. In both cases, by focusing the attention of the human expert on potential classification mistakes which are the most critical parts of the delineation, we reduce the amount of required supervision. We demonstrate the effectiveness of our approach on microscopy images depicting blood vessels and neurons.

Joint Segmentation and Path Classification of Curvilinear Structures

“…The simplest way to do this is to look for a minimum spanning tree [27], [28], [29], which can then be pruned. A more sophisticated approach that makes it possible to impose global geometric and topological constraints on the final delineation and allow loopy structures is to formulate the search for the optimal subgraph as Linear or Quadratic program [5], [30]. A recent trend is to also use two separate deep networks for this purpose.…”

Tracing in 2D to reduce the annotation effort for 3D deep delineation of linear structures

“…Delineation is a broad research topic. It operates on structures as different as roads (Mattyusand et al, 2017, Mnih, 2013, Mnih and Hinton, 2010, Wegner et al, 2013, blood vessels (Ganin andLempitsky, 2014, Maninis et al, 2016), bronchi (Meng et al, 2017), neurites (Peng et al, 2017, Sironi et al, 2016, and cell membranes (Mosinska et al, 2017), imaged using many different modalities. In this paper, we specifically address 3D delineation where the input is a volume, as opposed to a collection of ordered, but unregistered slices (Funke et al, 2012).…”

MetaBox+: A new Region Based Active Learning Method for Semantic Segmentation using Priority Maps

“…[32]. First AL approaches (before the deep learning (DL) era) for semantic segmentation, for instance based on conditional random fields, go back to [36,18,23,16]. At the heart of an AL method is the so-called query strategy that decides which data to present to the annotator / oracle next.…”