View article

Many different imaging modalities now routinely produce huge amounts of data thanks to increased acquisition speeds and extensive automation. Whilst computational image analysis techniques are constantly improving, it is often necessary to use manual analysis to some degree, in particular to provide ground truth annotations in order to train machine learning systems such as convolutional neural networks.

Volume electron microscopy techniques, such as serial block face SEM (SBF SEM), focused ion beam SEM (FIB SEM) and array tomography (AT)[1], produce datasets in the terabyte regime. In many cases, much of the analysis is still performed manually, which represents a major bottleneck in the workflow, restricting the amount of information that can be extracted from these rich datasets. Furthermore, single manual annotations may be prone to subjectivity that is not meaningfully encoded in the binary …