Machine learning as a tool for classifying electron tomographic reconstructions

Abstract

Electron tomographic reconstructions often contain artefacts from sources such as noise in the projections and a “missing wedge” of projection angles which can hamper quantitative analysis. We present a machine-learning approach using freely available software for analysing imperfect reconstructions to be used in place of the more traditional thresholding based on grey-level technique and show that a properly trained image classifier can achieve manual levels of accuracy even on heavily artefacted data, though if multiple reconstructions are being processed, a separate classifier will need to be trained on each reconstruction for maximum accuracy.