4.7 Article

Evaluation of Deep Learning Architectures for Complex Immunofluorescence Nuclear Image Segmentation

Journal

IEEE TRANSACTIONS ON MEDICAL IMAGING
Volume 40, Issue 7, Pages 1934-1949

Publisher

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TMI.2021.3069558

Keywords

Image segmentation; Computer architecture; Deep learning; Microscopy; Microprocessors; Training; Task analysis; Architecture evaluation; artificial images; deep learning; expert-annotated data; nuclear image segmentation

Funding

  1. Austrian Research Promotion Agency (FFG) COIN Networks project TISQUANT
  2. Austrian Science Fund (FWF) [I4162]
  3. Federal Ministry Republic of Austria for Climate Action, Environment, Energy, Mobility, Innovation and Technology (BMK)
  4. Federal Ministry Republic of Austria for Digital and Economic Affairs (BMDW)
  5. Province of Upper Austria
  6. Austrian Research Promotion Agency (FFG) COIN Networks project VISIOMICS
  7. Austrian Science Fund (FWF) [I4162] Funding Source: Austrian Science Fund (FWF)

Ask authors/readers for more resources

The study evaluates multiple deep learning architectures and conventional algorithms for complex fluorescence nuclear image segmentation, and introduces a novel strategy for creating artificial images. It shows that instance-aware segmentation architectures and Cellpose outperform U-Net architectures and conventional methods in terms of F1 scores, while U-Net architectures achieve higher mean Dice scores overall. Training with artificially generated images improves recall and F1 scores for complex images.
Separating and labeling each nuclear instance (instance-aware segmentation) is the key challenge in nuclear image segmentation. Deep Convolutional Neural Networks have been demonstrated to solve nuclear image segmentation tasks across different imaging modalities, but a systematic comparison on complex immunofluorescence images has not been performed. Deep learning based segmentation requires annotated datasets for training, but annotated fluorescence nuclear image datasets are rare and of limited size and complexity. In this work, we evaluate and compare the segmentation effectiveness of multiple deep learning architectures (U-Net, U-Net ResNet, Cellpose, Mask R-CNN, KG instance segmentation) and two conventional algorithms (Iterative h-min based watershed, Attributed relational graphs) on complex fluorescence nuclear images of various types. We propose and evaluate a novel strategy to create artificial images to extend the training set. Results show that instance-aware segmentation architectures and Cellpose outperform the U-Net architectures and conventional methods on complex images in terms of F1 scores, while the U-Net architectures achieve overall higher mean Dice scores. Training with additional artificially generated images improves recall and F1 scores for complex images, thereby leading to top F1 scores for three out of five sample preparation types. Mask R-CNN trained on artificial images achieves the overall highest F1 score on complex images of similar conditions to the training set images while Cellpose achieves the overall highest F1 score on complex images of new imaging conditions. We provide quantitative results demonstrating that images annotated by under-graduates are sufficient for training instance-aware segmentation architectures to efficiently segment complex fluorescence nuclear images.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.7
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available