Minimizing Manual Image Segmentation Turn-Around Time for Neuronal Reconstruction by Embracing Uncertainty

Stephen M Plaza,Mathew Saunders,Louis K Scheffer,Stefan Strack

doi:10.1371/journal.pone.0044448

Abstract

The ability to automatically segment an image into distinct regions is a critical aspect in many visual processing applications. Because inaccuracies often exist in automatic segmentation, manual segmentation is necessary in some application domains to correct mistakes, such as required in the reconstruction of neuronal processes from microscopic images. The goal of the automated segmentation tool is traditionally to produce the highest-quality segmentation, where quality is measured by the similarity to actual ground truth, so as to minimize the volume of manual correction necessary. Manual correction is generally orders-of-magnitude more time consuming than automated segmentation, often making handling large images intractable. Therefore, we propose a more relevant goal: minimizing the turn-around time of automated/manual segmentation while attaining a level of similarity with ground truth. It is not always necessary to inspect every aspect of an image to generate a useful segmentation. As such, we propose a strategy to guide manual segmentation to the most uncertain parts of segmentation. Our contributions include 1) a probabilistic measure that evaluates segmentation without ground truth and 2) a methodology that leverages these probabilistic measures to significantly reduce manual correction while maintaining segmentation quality.

Highlights

The proper segmentation of an image can facilitate analysis useful in many applications
We demonstrate the effectiveness of Generalized Probabilistic Rand Index (GPR) to guide manual segmentation refinement
The manual reconstruction is simulated by software that labels each edge yes/no based on the ground truth

Summary

Introduction

The proper segmentation of an image can facilitate analysis useful in many applications. Because small errors in the segmentation can result in large topological errors, manual inspection of the entire volume is necessary to correct any errors [1]. Results rely on the inspection of the entire image volume, even if automatic segmentation is used as the starting point. This need to look at every pixel forms a lower bound on the reconstruction effort, even if no errors are discovered or corrected. The quality of the segmentation algorithm is determined by measuring its similarity to ground truth. (The straightforward implementation of the Rand index takes quadratic time This algorithm is reduced to linear complexity by assigning each pixel in both images to a unique bin determined by the segmentation partition it belongs to. Pair-wise similarity is determined by determining the cardinality of disagreements of image A compared to image B’s partitions and image B compared to image A’s partitions.) In the following paragraphs, a brief description of the Rand Index (RI) [7] is provided

Methods

Results

Conclusion