Abstract
Marker-based alignment widely used for tilt series alignment in electron tomography (ET) is crucial to high-resolution tomographic reconstruction. However, accurate alignment with markers remains a challenge because it is difficult to detect markers accurately and obtain the precise positions of fiducial markers in the tilt series. Conventional marker detection algorithms highly depending on marker template and threshold for classification lack the adaptation for different types of samples. The classification accuracy is severely affected by high contrast structures other than markers and high-density areas. In this paper, we present an automatic fiducial marker detection algorithm that applies a fine-tuned classification model to fit with the particular dataset. The classification via a convolutional neural network (CNN) for marker detection is solved as a binary classification problem distinguishing between the positive samples and negative samples. Also, we established the training data for the model to learn the patterns of the fiducial marker and background noise. The experimental results indicate that our deep learning based marker detection algorithm can identify sufficient fiducial markers with high accuracy in a fully automatic manner and shows superiority compared with previous work.
Original language | English (US) |
---|---|
Title of host publication | 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) |
Publisher | Institute of Electrical and Electronics Engineers (IEEE) |
Pages | 642-645 |
Number of pages | 4 |
ISBN (Print) | 9781538654880 |
DOIs | |
State | Published - Feb 28 2019 |
Bibliographical note
KAUST Repository Item: Exported on 2020-10-01Acknowledgements: This research is supported by the National Key Research and Development Program of China (No. 2017YFA0504702 and 2017YFE0103900), the NSFC projects Grant (No. U1611263, U1611261, 61472397, 61502455 and 61672493) and Special Program for Applied Research on Super Computation of the NSFC-Guangdong Joint Fund (the second phase).