Cell Segmentation in Multi-modality High-Resolution Microscopy Images with Cellpose

Kwanyoung Lee, Hyungjo Byun, Hyunjung Shim
Proceedings of The Cell Segmentation Challenge in Multi-modality High-Resolution Microscopy Images, PMLR 212:1-11, 2023.

Abstract

Deep learning has achieved significant improvement in cell segmentation of microscopy images in the field of Biology. However, a lack of generalization has been a major bottleneck of segmentation models since the performance is largely degraded with out-of-distribution data or unseen class data. Developing a generalized segmentation model is challenging due to the diversity of modalities, different staining methods, complicated cell shapes, and extremely high image resolution in microscopy images. The dataset for the ‘’Weakly Supervised Cell Segmentation in Multi-modality High-Resolution Microscopy Images” challenge consists of images with these diverse characteristics. To address these challenges, we trained the Cellpose model to competently perform instance segmentation on datasets with various characteristics. For that, we 1) specified the model to only use green and blue channels for all types of cell images, and 2) investigated the effect and performance of the existing diameter estimation model to determine the areas where it performs best, using images of various resolutions. As a result, we achieved an F1 score of 0.7607 for the validation (Tuning) set.

Cite this Paper

BibTeX
@InProceedings{pmlr-v212-lee23b, title = {Cell Segmentation in Multi-modality High-Resolution Microscopy Images with Cellpose}, author = {Lee, Kwanyoung and Byun, Hyungjo and Shim, Hyunjung}, booktitle = {Proceedings of The Cell Segmentation Challenge in Multi-modality High-Resolution Microscopy Images}, pages = {1--11}, year = {2023}, editor = {}, volume = {212}, series = {Proceedings of Machine Learning Research}, month = {28 Nov--09 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v212/lee23b/lee23b.pdf}, url = {https://proceedings.mlr.press/v212/lee23b.html}, abstract = {Deep learning has achieved significant improvement in cell segmentation of microscopy images in the field of Biology. However, a lack of generalization has been a major bottleneck of segmentation models since the performance is largely degraded with out-of-distribution data or unseen class data. Developing a generalized segmentation model is challenging due to the diversity of modalities, different staining methods, complicated cell shapes, and extremely high image resolution in microscopy images. The dataset for the ‘’Weakly Supervised Cell Segmentation in Multi-modality High-Resolution Microscopy Images” challenge consists of images with these diverse characteristics. To address these challenges, we trained the Cellpose model to competently perform instance segmentation on datasets with various characteristics. For that, we 1) specified the model to only use green and blue channels for all types of cell images, and 2) investigated the effect and performance of the existing diameter estimation model to determine the areas where it performs best, using images of various resolutions. As a result, we achieved an F1 score of 0.7607 for the validation (Tuning) set.} }
Endnote
%0 Conference Paper %T Cell Segmentation in Multi-modality High-Resolution Microscopy Images with Cellpose %A Kwanyoung Lee %A Hyungjo Byun %A Hyunjung Shim %B Proceedings of The Cell Segmentation Challenge in Multi-modality High-Resolution Microscopy Images %C Proceedings of Machine Learning Research %D 2023 %E %F pmlr-v212-lee23b %I PMLR %P 1--11 %U https://proceedings.mlr.press/v212/lee23b.html %V 212 %X Deep learning has achieved significant improvement in cell segmentation of microscopy images in the field of Biology. However, a lack of generalization has been a major bottleneck of segmentation models since the performance is largely degraded with out-of-distribution data or unseen class data. Developing a generalized segmentation model is challenging due to the diversity of modalities, different staining methods, complicated cell shapes, and extremely high image resolution in microscopy images. The dataset for the ‘’Weakly Supervised Cell Segmentation in Multi-modality High-Resolution Microscopy Images” challenge consists of images with these diverse characteristics. To address these challenges, we trained the Cellpose model to competently perform instance segmentation on datasets with various characteristics. For that, we 1) specified the model to only use green and blue channels for all types of cell images, and 2) investigated the effect and performance of the existing diameter estimation model to determine the areas where it performs best, using images of various resolutions. As a result, we achieved an F1 score of 0.7607 for the validation (Tuning) set.
APA
Lee, K., Byun, H. & Shim, H.. (2023). Cell Segmentation in Multi-modality High-Resolution Microscopy Images with Cellpose. Proceedings of The Cell Segmentation Challenge in Multi-modality High-Resolution Microscopy Images, in Proceedings of Machine Learning Research 212:1-11 Available from https://proceedings.mlr.press/v212/lee23b.html.

Related Material