Improved cell segmentation using deep learning in label-free optical microscopy images

Schematic of the modified U-Net: a pretrained ResNet-18 encoder, residual blocks in the decoder, and a redesigned skip-connection. This is a schematic, not the paper's figure.

Abstract

The recently popular deep neural networks (DNNs) have a significant effect on the improvement of segmentation accuracy from various perspectives, including robustness and completeness in comparison to conventional methods. We determined that the naive U-Net has some lacks in specific perspectives and there is high potential for further enhancements on the model. Therefore, we employed some modifications in different folds of the U-Net to overcome this problem. Based on the probable opportunity for improvement, we develop a novel architecture by using an alternative feature extractor in the encoder of U-Net and replacing the plain blocks with residual blocks in the decoder. This alteration makes the model superconvergent yielding improved performance results on two challenging optical microscopy image series: a phase-contrast dataset of our own (MDA-MB-231) and a brightfield dataset from a well-known challenge (DSB2018). We utilized the U-Net with pretrained ResNet-18 as the encoder for the segmentation task. Hence, following the modifications, we redesign a novel skip-connection to reduce the semantic gap between the encoder and the decoder. The proposed skip-connection increases the accuracy of the model on both datasets. The proposed segmentation approach results in Jaccard Index values of 85.0% and 89.2% on the DSB2018 and MDA-MB-231 datasets, respectively. The results reveal that our method achieves competitive results compared to the state-of-the-art approaches and surpasses the performance of baseline approaches

Framework

Alternative encoder

A pretrained ResNet-18 replaces the plain U-Net encoder as the feature extractor.

Residual decoder

Plain blocks in the decoder are replaced with residual blocks.

Redesigned skip-connection

A new skip-connection reduces the semantic gap between encoder and decoder.

Two datasets

Evaluated on a phase-contrast MDA-MB-231 dataset and the brightfield DSB2018 challenge dataset, reaching Jaccard Index values of 89.2% and 85.0%.

Resources

Journal: TÜBİTAK journal page
DOI: 10.3906/elk-2105-244

BibTeX

@article{ayanzadeh2021improved,
  title = {Improved cell segmentation using deep learning in label-free optical microscopy images},
  author = {Ayanzadeh, Aydin and {\"O}zuysal, {\"O}zden Yal{\c{c}}in and Okvur, Devrim Pesen and {\"O}nal, Sevgi and T{\"o}reyin, Beh{\c{c}}et U{\u{g}}ur and {\"U}nay, Devrim},
  journal = {Turkish Journal of Electrical Engineering and Computer Sciences},
  volume = {29},
  number = {8},
  pages = {2855--2868},
  year = {2021},
  doi = {10.3906/elk-2105-244}
}