[Paper Review] BriFiSeg: a deep learning-based method for semantic and instance segmentation of nuclei in brightfield images

Gendarme Mathieu, Lambert Annika M.+1 more|arXiv (Cornell University)|Nov 6, 2022

Cell Image Analysis Techniques4 citations

TL;DR

BriFiSeg is a deep learning framework that enables accurate semantic and instance segmentation of nuclei in unstained brightfield microscopy images using U-Net-based architectures with pre-trained encoders. It achieves robust performance across diverse biological contexts, including treated cell lines, without requiring immunohistochemical staining or specialized sample preparation.

ABSTRACT

Generally, microscopy image analysis in biology relies on the segmentation of individual nuclei, using a dedicated stained image, to identify individual cells. However stained nuclei have drawbacks like the need for sample preparation, and specific equipment on the microscope but most importantly, and as it is in most cases, the nuclear stain is not relevant to the biological questions of interest but is solely used for the segmentation task. In this study, we used non-stained brightfield images for nuclei segmentation with the advantage that they can be acquired on any microscope from both live or fixed samples and do not necessitate specific sample preparation. Nuclei semantic segmentation from brightfield images was obtained, on four distinct cell lines with U-Net-based architectures. We tested systematically deep pre-trained encoders to identify the best performing in combination with the different neural network architectures used. Additionally, two distinct and effective strategies were employed for instance segmentation, followed by thorough instance evaluation. We obtained effective semantic and instance segmentation of nuclei in brightfield images from standard test sets as well as from very diverse biological contexts triggered upon treatment with various small molecule inhibitor. The code used in this study was made public to allow further use by the community.

Motivation & Objective

• To develop a method for nuclei segmentation in brightfield microscopy images without relying on nuclear stains.
• To eliminate the need for specialized sample preparation and immunohistochemical staining in cell segmentation tasks.
• To enable segmentation on live or fixed samples using standard brightfield microscopes.
• To evaluate the performance of pre-trained encoders in combination with U-Net architectures for nuclei segmentation.
• To provide a publicly available, generalizable solution for nuclei segmentation applicable across various biological contexts.

Proposed method

• Utilizes U-Net-based deep neural network architectures for semantic segmentation of nuclei in brightfield images.
• Employs pre-trained encoders (e.g., ResNet, EfficientNet) to improve feature extraction and model generalization.
• Applies two distinct strategies for instance segmentation, enabling separation of individual nuclei from overlapping or clustered regions.
• Systematically evaluates different encoder architectures to identify optimal combinations with the segmentation head.
• Trains and validates models on four distinct cell lines, including those treated with small molecule inhibitors.
• Makes the source code publicly available to support community adoption and further development.

Experimental results

Research questions

• RQ1 Can deep learning models achieve accurate nuclei segmentation in unstained brightfield images without reliance on nuclear stains?
• RQ2 How do pre-trained encoders impact the performance of U-Net-based segmentation models in brightfield microscopy?
• RQ3 What are the most effective strategies for instance segmentation in dense or overlapping nuclear regions in brightfield images?
• RQ4 Can the method generalize across diverse biological conditions, including drug-treated samples?
• RQ5 How does the performance of BriFiSeg compare to existing stain-dependent segmentation methods in terms of accuracy and robustness?

Key findings

• BriFiSeg achieves effective semantic and instance segmentation of nuclei in brightfield images across four distinct cell lines.
• The method demonstrates robust performance on both standard test sets and biologically diverse contexts, including samples treated with small molecule inhibitors.
• Pre-trained encoders significantly improve segmentation performance when combined with U-Net architectures.
• Two distinct instance segmentation strategies were found to be effective, enabling accurate separation of individual nuclei.
• The framework is applicable to both live and fixed samples, requiring no specialized staining or sample preparation.
• The codebase is publicly released, enabling reproducibility and broader adoption in the research community.