State-of-the-art nuclei segmentation techniques using deep learning technology solve many problems related to extracting reliable quantitative information on the cell nucleus level.
This article presents Deep Learning (DL) based methods used for nuclei segmentation tasks. Learn about the most common challenges researchers face during tissue analysis like over- and under-segmentation, overlapping or touching nuclei, and domain shift.
We are here to acquaint you with preprocessing, annotation, and labeling practices. This advice will assist you in achieving the best possible results when incorporating nuclei segmentation into your digital image analysis pipeline.
Find out about the various uses of application training for nuclei segmentation tasks in histopathology.
- How automated image analysis will help you advance in nuclei segmentation methods?
- The cell nuclei segmentation workflow
- Analysis frameworks, challenges, and opportunities in nuclei segmentation methods
- Understanding the benefits of deep learning-based nuclei segmentation techniques in histopathology image analysis
- A simple nuclei segmentation approach using IKOSA AI
How automated image analysis will help you advance in nuclei segmentation methods?
The popularity of complex methods for nuclei segmentation using DL has significantly increased. These approaches actively focus on automating the detection and measurement of nuclei within tissue sections through sophisticated image recognition models.
Accurately detecting and segmenting cell nuclei in microscopy images is a vital goal across diverse disciplines, from cytology to histology and general pathology. Many view changes in the morphological attributes of nuclei within tissue samples as indicators of pathological processes.
Yet, the choice of the right AI models and software tools for your research project should depend primarily on their performance in terms of processing efficiency and segmentation accuracy.
The cell nuclei segmentation workflow
The automated segmentation of cell nuclei involves several stages: preprocessing, segmentation, postprocessing, and evaluation.
- During the preprocessing stage, the quality of input images is improved to ensure optimal segmentation results.
- The objective of the actual segmentation stage is to extract cell nuclei present in the image foreground.
- During the post-processing stage, the segmentation results are optimized. It involves morphological operations and filtering, such as sorting out false findings (false positives) from the segmentation results. Post-processing can also be applied to refine nuclear boundaries and set size thresholds.
- At the evaluation stage, the researcher assesses the quality of the quantitative segmentation results based on benchmark metrics. The most common metrics used at this stage are precision, recall, Intersection over Union (IoU), Average Precision (AP), Dice Coefficient (DC), and F1-measure (Win et al., 2018).
Precision is a measure of prediction accuracy. Recall is a metric related to the reproducibility of results. The F1 measure is a metric related to the object detection capabilities of the network. All the metrics mentioned above can express values from 0 to 1. The closer the value is to 1, the higher the predictive model power is (Fujita & Han, 2020). IoU stands for the intersection over the union of nuclei areas on a ground truth mask and the output of the proposed model (Chen et al. 2019). The Dice Coefficient is a quality metric used for the pixel-by-pixel comparison of the ground truth with the segmentation results. DC also plays a central role in assigning pixels to categories such as true positives, false positives, and false negatives (Jung et al. 2019).
IKOSA AI is your research partner, ready to assist you.
Analysis frameworks, challenges, and opportunities in nuclei segmentation methods
A frequent and central goal of tissue analysis is determining the population of individual cell nuclei in microscopy images. Based on the task, object detection models may be sufficient to localize nuclei instances.
Morphological properties of the nucleus, including shape and texture, enable the extraction of quantitative details at the level of individual objects. This data can then be employed to visualize size and distribution as well as to identify abnormal patterns.
A crucial phase within this analysis workflow involves precisely dividing the image into semantic classes. These classes denote cohesive regions, such as tissue types (e.g. epithelial and stroma), cell types (e.g. neurons), cellular structures (e.g. nuclei, cytoplasm), and subcellular compartments within a histology dataset.
It is essential to detect each semantic class by accurately identifying the boundaries of individual nuclei, a task known as boundary detection. As a result, the nuclear segmentation process categorizes pixels into three distinct types: background, nuclei interior, and nuclear boundaries (Caicedo et al., 2019).
Tips and tricks
Using state-of-the-art deep learning techniques enables you to detect individual instances of a particular type of nuclei.
The challenge of traditional nuclei segmentation methods
The most common traditional approaches to nuclei segmentation analysis are thresholding and seeded watersheds.
Thresholding utilizes intensity values from grayscale histograms of microscopy images to filter objects of interest. This method is particularly effective for extracting nuclei from the foreground (Win et al., 2018; Caicedo et al., 2019).
The classical seeded watershed method divides microscope images by employing watershed lines aligned with the boundaries of morphological structures like nuclei. This process identifies the positions of nuclei, known as seeds. The AI model then utilizes region growing to progressively expand each seed until the nucleus boundaries are encompassed (Atta-Fosu et al., 2016).
However, accurate and robust nuclear instance segmentation using traditional methods can be challenging due to morphological variations in the nuclei among different organs and tissue types. Moreover, spatial configurations (e.g. clusters, touching nuclei), tissue preparation specifics, or complex backgrounds containing imaging artifacts (e.g. folds, out-of-focus regions) limit the applicability of such methods (Zhou et al., 2019).
Simultaneously, there has been a surge in innovative deep learning-based methods that have demonstrated enhanced efficiency and robustness in segmenting cell nuclei across diverse tissue types (Jang et al., 2019). Here, we present a concise overview of the available DL-based approaches to nuclei segmentation.
Deep learning models for nuclei segmentation
Deep neural network models apply semantic and instance segmentation to identify and locate nuclei in microscopy images (Naylor et al., 2017).
Most common DL-based approaches for nuclei segmentation:
Semantic segmentation involves assigning areas of the image to different semantic classes such as nuclei area, cytoplasm area, nuclei boundaries, and background area. In other words, it is a method used for labeling each pixel in an image with the corresponding class that it belongs to. The semantic segmentation result is a segmentation map in which each pixel is colored or labeled based on its predicted class. This map provides a detailed representation of the different objects and structures present in the image. (Cui et al., 2019; Kowal et al., 2020).
Instance segmentation comprises two distinct stages: semantic segmentation and object identification. This advanced computer vision technique surpasses the capabilities of semantic segmentation. While semantic segmentation focuses on labeling pixels with semantic classes, instance segmentation takes analysis a step further. It doesn’t only assign class labels but also distinguishes individual instances of objects within those classes. In other words, instance segmentation differentiates between multiple objects of the same class and provides a unique label or identifier for each instance.
Did you know?
Instance segmentation offers the advantage of providing additional information on rich morphological features of cellular structures.
DL models classify and segment objects in the images by learning features from large amounts of representative input data. (Hayakawa et al., 2019). Various types of DL architectures differing in layer configuration and model depth have found use in segmenting biomedical images (Kowal et al., 2020).
Create your segmentation application without a line of code.
These are the most common DL architectures used for nuclei segmentation:
Fully convolutional neural networks (FCN)
The fully convolutional network (FCN) approach relies on the original CNN architecture but uses upsampling and skip layers.
FCNs operate on image input of arbitrary size and, instead of single pixel classifications, generate entire images as output (semantic maps). The spatial precision of the output is more refined, and the network is more sensitive to variations in input image resolution. It makes FCN flexible enough not to require extensive pre- and post-processing and to learn high-level features (Zhang et al., 2017; Naylor et al., 2017; Long et al., 2015).
FCNs are known to be particularly precise in capturing deep features of cell nuclei, but not as effective in localizing nucleus boundaries. This type of architecture has been quoted for the capability to segment “normal” and “abnormal” nucleus masks in cytology studies (Zhang et al., 2017).
U-net
U-Net is a U-shaped multi-scale stacked neural network based on FCN architecture. A U-Net relies on the stacking of encoder and decoder layers. The layer stacking allows the model to learn more complex features as network depth increases (Vuola et al., 2019).
Encoder layers create representation-encoding by reducing input images by their spatial resolution through multiple convolutional layers. Decoder layers create feature maps to predict segmentation masks.
Similarly to the FCN, a typical U-Net introduces shortcut connections between the encoder and decoder layers. These skip connectors allow the network to retrieve spatial information lost during the process of pooling and to preserve features at different resolutions. That is why U-Net is particularly effective when dealing with a small training dataset, as this framework is flexible as regards training data availability and problem complexity. (Ronneberger et al., 2015; Kang et al., 2019; Vuola et al., 2019; Ibtehaz & Rahman, 2020).
U-Net provides excellent results in nucleus segmentation tasks, especially in large nuclei, elliptical nuclei shapes, and single nuclei. Certain drawbacks of U-Net in comparison to other networks can still be observed in tasks related to nucleus detection and multiple nuclei merging into a single larger nucleus (Vuola et al., 2019).
MASK R CNN
MASK R-CNN is an example of a domain adaptation network, particularly effective in instance segmentation. Mask R-CNN enables the simultaneous detection and segmentation of cell images by adding branching subnets to estimate the segmentation mask for each region of interest (ROI) (Allehaibi et al., 2019; Fujita & Han, 2020). First, the model identifies the bounding boxes of nuclei, followed by conducting segmentation exclusively within these boxes for the respective nuclei (Vuola et al., 2019).
When comparing histopathological image inputs with the nucleus segmentation mask generated by different DL models, distinct discrepancies may arise, primarily concerning the assignment of objects to various classes. Due to such image-level and feature-level domain bias, a phenomenon referred to as “domain shift” occurs. Domain adaptation deep learning methods tackle this issue by adding benchmark datasets as a source to adapt to a diverse range of biomedical images (Lui et al., 2020; Hsu et al., 2021).
Mask R-CNN performs outstandingly in nucleus detection tasks, while its segmentation capabilities still need improvement. Mask R-CNN has been observed to perform better in nuclei detection, but less effectively in nuclei segmentation in comparison to U-Net. As opposed to U-Net, this specific deep learning model produces fewer instances of over-segmentation and demonstrates improved accuracy in detecting clumped or grouped nuclei. Still, Mask R-CNN tends to over-segment large nuclei, while it yields better performance for small and middle-sized nuclei (Vuola et al., 2019).
Each type of deep learning network has its advantages and is particularly efficient at different nucleus segmentation tasks. Some authors even suggest that a combination of two AI models, e.g. U-Net and Mask R-CNN, contributes to improved efficiency in segmenting the nuclei of individual cells (Vuola et al. 2019). Choosing the right network for your research project or adapting a deep learning pipeline to match your requirements requires expert knowledge in computer vision and programming skills.
Tips and tricks
When selecting the right model for your nucleus segmentation project, you should choose one that is robust and flexible enough to adapt to different samples, research designs, and segmentation tasks.
Essential steps for preparing training data
Before conducting the actual quantitative segmentation analysis, one extra stage needs to be performed to prepare the image dataset for app training.
Training data selection and preparation
Depending on the complexity of the data, DL models can be effective in segmenting nuclei even when trained on a small dataset. Yet, providing a variety of training images is recommended as it improves the app’s predictive power. A DL method can be used across different experiments as long as variations of the morphological structures of interest have been previously introduced to the model (Caicedo et al. 2019). Otherwise, a “domain shift” issue can occur (Liu et al., 2020; Hsu et al., 2021).
Don’t forget to pay attention to the input images you choose. This helps achieve a balanced dataset by reducing the number of images with background only or subpar quality (Araujo et al., 2019).
Image data augmentation
Augmentation refers to the artificial extension of a dataset to increase the amount and variance of training data so that it best represents the expected target domain. Data augmentation is amongst the most important steps in preprocessing for building a robust model and greater accuracy.
Data augmentation involves applying several transformations or modifications to the original image dataset to create new training examples. These transformations introduce realistic variations that the model might encounter in real-world scenarios. Through training on both the original and augmented data, the model enhances its capability to identify and manage multiple image variations effectively.
Common augmentation strategies involve random or parameterized spatial transformations such as flipping, rotation, or deformation, frequently combined with adding noise or color augmentation (Allehaibi et al., 2019). A nuclei segmentation model generally benefits from these operations when applied correctly. By employing meaningful data augmentation, the total amount of manually annotated and labeled data performed before training can be kept at a minimum, facilitating faster model development.
Color processing and normalization
During the segmentation process of stained images, color information is an important indicator for detecting objects of interest. Nevertheless, challenges can arise for DL models in properly processing the data when variations in staining or acquisition devices have been used for the different whole-slide images.
Certain color variations may occur when using different staining protocols, stain brands, or microscope and scanning devices. These can impact the performance of the deep learning network, mainly when discrepancies in color hues exist between the training dataset and the images used for analysis (Kowal et al., 2020).
Consequently, if you train the model on a homogeneous dataset, its analysis capabilities might be limited to data that aligns with a specific color scheme. When your experiment requires variations in color and texture, it becomes crucial to guarantee the training of a robust model. You can achieve it by incorporating training data that covers the complete spectrum of images you intend to analyze.
Annotation and labeling strategies
Clarity and organization are of utmost importance to good annotation practice. To begin with, it’s crucial to define your annotation classes clearly, creating a well-structured hierarchy. Employing Regions of Interest (ROIs) helps to precisely enclose structures and focus the analysis on particular areas containing structures of interest.
Consistency is key, both in applying color schemes for different classes and in marking annotations with distinct labels. The number of annotations used depends on analysis goals, feature complexity, resources available, and model requirements, typically involving a representative sample for each class.
An iterative approach to annotation is advisable, allowing for incremental increases based on model performance and analysis needs. Diversity in annotated images and consideration of object variability, especially in shape, size, and appearance, is essential for robust model performance.
Don’t forget to pay attention to the background of images, where noise and artifacts can often be observed. By annotating these regions, your model can learn to distinguish between actual objects and anomalies. Precision in outlining boundaries and consistent annotation styles maintain uniformity and reduce confusion.
Balancing data volume, annotation quality, and analysis requirements are the main goals. Properly performed annotation work, tailored to your analysis goals and the nature of your objects of interest, ensures reliable and meaningful results.
In another article, we comprehensively explore the fundamentals of image annotation alongside more advanced techniques. We also demonstrate how effortless it is to annotate digital images using the IKOSA software.
Upload your images, annotate them and train your unique app.
Dealing with touching and overlapping nuclei during app development
The reliable and accurate detection of overlapping, touching, and heavily clustered nuclei poses a major challenge when training automated nuclei segmentation models with the majority of solutions available on the market.
Some software solutions may provide inaccurate identification of nuclear boundaries due to either under-segmentation (detecting multiple nuclei as one) or over-segmentation (detecting one single nucleus as multiple ones (Mahmood et al., 2019; Zhou et al., 2019).
Achieving optimal segmentation results requires choosing appropriate segmentation techniques, fine-tuning parameters, and post-processing methods to correct any errors that arise from under-segmentation or over-segmentation.
Several studies have attempted to address this issue using non-DL methods, such as watershed transform algorithms (Cloppet & Boucher, 2008) or active shape models (ASM) (Plissiti & Nikou, 2012). DL-based models for splitting touching and overlapping nuclei areas are more scarce and often involve complementing the DL algorithm with traditional techniques like seeded watershed or post-processing procedures (Kowal et al., 2020).
Instance segmentation is often cited as a reliable approach for separating touching or overlapping nuclei. This method is particularly robust in such cases due to the availability of a contour-aware network and sensitivity to spatial and textural dependencies between cell nuclei (Zhou et al., 2019).
Tips and tricks
Use the instance segmentation method to minimize the issue of over- or under-segmentation of the nucleus instances in your dataset.
Understanding the benefits of deep learning-based nuclei segmentation techniques in histopathology image analysis
Let’s delve deeper into the various uses of the DL-based nucleus segmentation method. In histopathology studies, nucleus segmentation is a pivotal task facilitating the analysis of nuclei morphology, classification of cell types, and critical functions like cancer detection and grading.
One major challenge that DL-based methods face during the analysis of histopathology image data is the segmenting of abnormal cell structures. Abnormalities in the morphological characteristics of the nuclei such as size, shape, texture, color, volume, and improper distribution of nuclear to cytoplasmic ratio serve as indicators of pathological processes detected in histology images (Ishrad et al., 2013; Rączkowski et al., 2019).
In histopathology studies, the number of cells undergoing mitosis, which is an indicator of cell division, is frequently used as an indicator of tumor severity. Precisely tracking instances of mitotic nuclei is crucial for this purpose. However, this task is challenging due to the variability in the morphology and texture of mitotic cells (Wang et al., 2014).
Moreover, histopathology studies have shown that the performance of trained models can vary significantly when applied to tissues from different organs (Kang et al., 2019; Cui et al., 2019). This variability underlines the pressing need for effective nuclei segmentation methods that can be applied consistently across various cell-, tissue-, and organ types.
Researchers dedicate a notable portion of histopathology nuclei segmentation studies to investigating two distinct nucleus types: lymphocyte nuclei and epithelial nuclei. These two types are frequently linked to inflammatory processes on a cellular level, thus forming a thematic focal point (Chen et al., 2018). Nuclei segmentation studies involving histopathological images have been conducted across various tissue types, with the most prevalent ones including breast, cervix, and prostate tissues.
A simple nuclei segmentation approach using IKOSA AI
In this section, we provide you with a practical example of how to train an application for nuclei segmentation using real-life data. By adhering to the proposed workflow and paying attention to the fundamental steps we previously discussed, you can improve and streamline your analysis experience.
Our primary focus is on the proper annotation process and the training of AI applications. If you are not already acquainted with the IKOSA platform and the steps for creating a project and uploading images, we recommend utilizing our knowledge base, which covers all these aspects in detail.
IKOSA AI allows you to quickly segment nuclei without wasting time on coding.
Materials and Methods
For this demonstration, we carefully selected 11 images that are representative of our mouse prostate dataset, including different background variations and several nuclei morphologies. Our objective was to assemble a set of training images that encompassed the full range of variations present within the entire dataset.
Before starting the annotation process, you must create labels you want to assign to your annotations. These labels essentially describe or categorize your annotated objects. In our case, we named the label “Nuclei”. We chose the freeform annotation tool as the most appropriate for nuclei annotation. Please ensure you annotate all the nuclei of interest within the image, or your selected Region of Interest (ROI), as applicable.
To narrow down the training to specific areas we used Regions of Interest (ROIs). Another advantage of utilizing ROIs is that you are not required to annotate the entire image, which can be a rather laborious task.
At the start of the application training process, we opted for instance segmentation as the appropriate training type. This method allows us to leverage object detection and effectively separate touching nuclei.
We chose a manual data split, which means that we divided the images into a training dataset (80%) and a validation dataset (20%) ourselves. But it’s also perfectly fine to employ a random data split where your images will be randomly allocated to the training and validation datasets.
Did you know?
The validation data serves as a crucial feature in assessing the performance of the trained application.
Having performed these initial steps, we commenced with a “quick training” session, allowing us a rapid preliminary assessment of the app’s performance.
This step enabled us to determine whether additional training data was necessary or if adjustments to annotations in the training set were required. For more refined results, we recommend employing the “extended training” option after optimizing the data. This typically results in significant improvements.
Did you know?
We advise you to choose a clear name for your project and provide a detailed description. This ensures that both you and other colleagues involved in the process are well-informed about the purpose of the application training.
With IKOSA AI, you can quickly solve complex nucleus segmentation problems.
Breaking down app training results
The “quick training” session finished in less than 15 minutes, after which IKOSA AI promptly sent an email notification upon completion. You can download the training report as a PDF file with a breakdown of quantitative and qualitative results.
The overall performance (Fig. 1) of the trained app seems quite okay considering the short training time. However, we were not happy with the result (Fig. 2) and saw the potential for optimization. As a result, we conducted a thorough review of all existing annotations and made necessary adjustments before starting with the “extended training” session. Moreover, we increased our annotation efforts to generate more input data, particularly focusing on objects that were not segmented adequately in previous iterations.
These optimization steps led to an overall improvement in the precision of the trained model, resulting in even more accurate nuclei segmentation, as you can see in Fig. 3.
Did you know?
Precise annotation of the data is crucial to ensuring accurate results. In general, if you are not satisfied with your app’s performance, there are several ways to improve your app training results.
Detected objects are compared to the ground-truth labels set by the user, resulting in a classification of true-positive, false-positive, and false-negative instances. Following this, the deep learning model is trained to identify morphological structures akin to the ground truth annotations.
However, if you are happy with the performance of your trained application and have carefully reviewed the report, you can deploy and use it for the analysis of your entire data set.
Analysis results
To fulfill the objectives of this use case and to conclude the workflow, we performed an analysis of unseen image data using the trained application.
The instance segmentation app provides a CSV file with information on the analyzed images and objects such as ROI ID, area, number of Nuclei, and many other parameters, such as circularity and eccentricity.
It’s important to keep in mind that striving for 100% perfect outcomes in image analysis may not be a realistic goal in many cases. Instead, the primary objective should be the continuous improvement of results through iterative testing and refinement until your model delivers reliable and reproducible results.
Conclusion
As you can probably see by now, training a nuclei segmentation app typically involves deep learning techniques and several steps to develop a model that can accurately identify and segment nuclei in images. IKOSA AI offers an easy and accessible approach to application training with no coding skills required. Moreover, our dedicated team remains at your disposal, providing unwavering user support and invaluable practical guidance throughout your journey.
Train your segmentation application in IKOSA without coding or AI experience!
Our authors:
Lead content writer, life science professional, and simply a passionate person about technology in healthcare
Content writer focused on the technological advancements in healthcare such as digital health literacy and telemedicine.
Health communications and marketing expert dedicated to delivering the latest topics in life science technology to healthcare professionals.
References
Ali, T., Masood, K., Irfan, M., Draz, U., Nagra, A. A., Asif, M., … & Yasin, S. (2020). Multistage Segmentation of Prostate Cancer Tissues Using Sample Entropy Texture Analysis. Entropy, 22(12), 1370.
Al-Kofahi, Y., Zaltsman, A., Graves, R., Marshall, W., & Rusu, M. (2018). A deep learning-based algorithm for 2-D cell segmentation in microscopy images. BMC bioinformatics, 19(1), 1-11.
Allehaibi, K. H. S., Nugroho, L. E., Lazuardi, L., Prabuwono, A. S., & Mantoro, T. (2019). Segmentation and classification of cervical cells using deep learning. IEEE Access, 7, 116925-116941.
Araujo, F. H., Silva, R. R., Ushizima, D. M., Rezende, M. T., Carneiro, C. M., Bianchi, A. G. C., & Medeiros, F. N. (2019). Deep learning for cell image segmentation and ranking. Computerized Medical Imaging and Graphics, 72, 13-21.
Alom, M. Z., Yakopcic, C., Taha, T. M., & Asari, V. K. (2018). Microscopic nuclei classification, segmentation, and detection with improved Deep Convolutional Neural Network (DCNN) approaches—arXiv preprint arXiv:1811.03447.
Bhattacharjee, S., Park, H. G., Kim, C. H., Prakash, D., Madusanka, N., So, J. H., … & Choi, H. K. (2019). Quantitative analysis of benign and malignant tumors in histopathology: Predicting prostate cancer grading using SVM. Applied Sciences, 9(15), 2969.
Bruch, R., Rudolf, R., Mikut, R., & Reischl, M. (2020). Evaluation of semi-supervised learning using sparse labeling to segment cell nuclei. Current Directions in Biomedical Engineering, 6(3), 398-401.
Caicedo, J. C., Roth, J., Goodman, A., Becker, T., Karhohs, K. W., Broisin, M., … & Carpenter, A. E. (2019). Evaluation of deep learning strategies for nucleus segmentation in fluorescence images. Cytometry Part A, 95(9), 952-965.
Chen, L., Deng, H., Cui, H., Fang, J., Zuo, Z., Deng, J., Li, Y., Wang, X., & Zhao, L. (2017). Inflammatory responses and inflammation-associated diseases in organs. Oncotarget, 9(6), 7204–7218. https://doi.org/10.18632/oncotarget.23208
Chen, K., Zhang, N., Powers, L., & Roveda, J. (2019, April). Cell nuclei detection and segmentation for computational pathology using deep learning. In 2019 Spring Simulation Conference (SpringSim) (pp. 1-6). IEEE.
Cloppet, F., & Boucher, A. (2008, December). Segmentation of overlapping/aggregating nuclei cells in biological images. In 2008 19th International Conference on Pattern Recognition (pp. 1-4). IEEE.
Cruz-Roa, A. A., Ovalle, J. E. A., Madabhushi, A., & Osorio, F. A. G. (2013, September). A deep learning architecture for image representation, visual interpretability, and automated basal-cell carcinoma cancer detection. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 403-410). Springer, Berlin, Heidelberg.
Cui, Y., Zhang, G., Liu, Z., Xiong, Z., & Hu, J. (2019). A deep learning algorithm for one-step contour aware nuclei segmentation of histopathology images. Medical & biological engineering & computing, 57(9), 2027-2043.
Fujita, S., & Han, X. H. (2020). Cell Detection and Segmentation in Microscopy Images with Improved Mask R-CNN. In ACCV Workshops (pp. 58-70).
Hayakawa, T., Prasath, V. S., Kawanaka, H., Aronow, B. J., & Tsuruoka, S. (2019). Computational nuclei segmentation methods in digital pathology: A survey. Archives of Computational Methods in Engineering, 1-13.
Ho, D. J., Yarlagadda, D. V., D’Alfonso, T. M., Hanna, M. G., Grabenstetter, A., Ntiamoah, P., … & Fuchs, T. J. (2021). Deep multi-magnification networks for multi-class breast cancer image segmentation. Computerized Medical Imaging and Graphics, 88, 101866.
Hsu, J., Chiu, W., & Yeung, S. (2021). DARCNN: Domain Adaptive Region-based Convolutional Neural Network for Unsupervised Instance Segmentation in Biomedical Images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 1003-1012).
Ibtehaz, N., & Rahman, M. S. (2020). MultiResUNet: Rethinking the U-Net architecture for multimodal biomedical image segmentation. Neural Networks, 121, 74-87.
Irshad, H., Veillard, A., Roux, L., & Racoceanu, D. (2013). Methods for nuclei detection, segmentation, and classification in digital histopathology: a review—current status and future potential. IEEE reviews in biomedical engineering, 7, 97-114.
Jung, H., Lodhi, B., & Kang, J. (2019). An automatic nuclei segmentation method based on deep convolutional neural networks for histopathology images. BMC Biomedical Engineering, 1(1), 1-12.
Kang, Q., Lao, Q., & Fevens, T. (2019, October). Nuclei segmentation in histopathological images using two-stage learning. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 703-711). Springer, Cham.
Kowal, M., Żejmo, M., Skobel, M., Korbicz, J., & Monczak, R. (2020). Cell nuclei segmentation in cytological images using convolutional neural network and seeded watershed algorithm. Journal of digital imaging, 33(1), 231-242.
Lee, J., Kim, H., Cho, H., Jo, Y., Song, Y., Ahn, D., … & Ye, S. J. (2019). Deep-learning-based label-free segmentation of cell nuclei in time-lapse refractive index tomograms. Ieee Access, 7, 83449-83460.
Liu, D., Zhang, D., Song, Y., Zhang, F., O’Donnell, L., Huang, H., … & Cai, W. (2020). Unsupervised instance segmentation in microscopy images via panoptic domain adaptation and task re-weighting. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 4243-4252).
Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 3431-3440).
Lu, Z., Xu, S., Shao, W., Wu, Y., Zhang, J., Han, Z., … & Huang, K. (2020). Deep-learning–based characterization of tumor-infiltrating lymphocytes in breast cancers from histopathology images and multiomics data. JCO clinical cancer informatics, 4, 480-490.
Mahmood, F., Borders, D., Chen, R. J., McKay, G. N., Salimian, K. J., Baras, A., & Durr, N. J. (2019). Deep adversarial training for multi-organ nuclei segmentation in histopathology images. IEEE transactions on medical imaging, 39(11), 3257-3267.
Naylor, P., Laé, M., Reyal, F., & Walter, T. (2017, April). Nuclei segmentation in histopathology images using deep neural networks. In 2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017) (pp. 933-936). IEEE.
Nguyen, K.; Jain, A.K.; Allen, R.L. Automated gland segmentation and classification for Gleason grading of prostate tissue images. In: Proceedings of the 20th International Conference on Pattern Recognition (ICPR), Istanbul, Turkey, 23–26 August 2010; Volume 1, pp. 1497–1500.
Plissiti, M. E., & Nikou, C. (2012). Overlapping cell nuclei segmentation using a spatially adaptive active physical model. IEEE Transactions on Image Processing, 21(11), 4568-4580.
Qu, H., Wu, P., Huang, Q., Yi, J., Yan, Z., Li, K., … & Metaxas, D. N. (2020). Weakly supervised deep nuclei segmentation using partial points annotation in histopathology images. IEEE Transactions on Medical Imaging, 39(11), 3655-3666.
Rączkowski, Ł., Możejko, M., Zambonelli, J., & Szczurek, E. (2019). ARA: accurate, reliable, and active histopathological image classification framework with Bayesian deep learning. Scientific reports, 9(1), 1-12.
Ren, J., Sadimin, E., Foran, D. J., & Qi, X. (2017, February). Computer-aided analysis of prostate histopathology images to support a refined Gleason grading system. In Medical Imaging 2017: Image Processing (Vol. 10133, p. 101331V). International Society for Optics and Photonics.
Ronneberger, O., Fischer, P., & Brox, T. (2015, October). U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical Image Computing and computer-assisted intervention (pp. 234-241). Springer, Cham.
Saha, M., Chakraborty, C., & Racoceanu, D. (2018). Efficient deep learning model for mitosis detection using breast histopathology images. Computerized Medical Imaging and Graphics, 64, 29-40.
Vuola, A. O., Akram, S. U., & Kannala, J. (2019, April). Mask-RCNN and U-net ensembled for nuclei segmentation. In 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019) (pp. 208-212). IEEE.
Wang, H., Xian, M., & Vakanski, A. (2020, April). Bending loss regularized network for nuclei segmentation in histopathology images. In 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI) (pp. 1-5). IEEE.
Wang, H., Roa, A. C., Basavanhally, A. N., Gilmore, H. L., Shih, N., Feldman, M., … & Madabhushi, A. (2014). Mitosis detection in breast cancer pathology images by combining handcrafted and convolutional neural network features. Journal of Medical Imaging, 1(3), 034003.
Win, K. Y., Choomchuay, S., Hamamoto, K., & Raveesunthornkiat, M. (2018). Comparative study on automated cell nuclei segmentation methods for cytology pleural effusion images. Journal of Healthcare Engineering, 2018.
Yoo, I., Yoo, D., & Paeng, K. (2019, October). PseudoEdgeNet: Nuclei segmentation only with point annotations. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 731-739). Springer, Cham.
Zhou, Y., Onder, O. F., Dou, Q., Tsougenis, E., Chen, H., & Heng, P. A. (2019, June). Cia-net: Robust nuclei instance segmentation with contour-aware information aggregation. In International Conference on Information Processing in Medical Imaging (pp. 682-693). Springer, Cham.
Zhang, L., Sonka, M., Lu, L., Summers, R. M., & Yao, J. (2017, April). Combining fully convolutional networks and graph-based approach for automated segmentation of cervical cell nuclei. In 2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017) (pp. 406-409). IEEE.
Zhang, Z., Leong, K. W., Van Vliet, K., Barbastathis, G., & Ravasio, A. (2021). Deep learning for label-free nuclei detection from implicit phase information of mesenchymal stem cells. Biomedical Optics Express, 12(3), 1683-1706.