Cell Segmentation Research Articles (Page 1)

PathViT: Automated disease classification from skeletal muscle histopathology.

Integration of single-cell and spatial transcriptomes represents a fundamental strategy to enhance spatial data quality. However, existing methods for mapping single-cell data to spatial coordinates struggle with large-scale data sets comprising millions of cells. Here, we introduce Polyomino, an intelligent region-allocation method inspired by the region-of-interest (ROI) concept from image processing. By using gradient descent, Polyomino allocates cells to structured spatial regions that match the most significant biological information, optimizing the integration of data and improving speed and accuracy. Polyomino excels in integrating data even in the presence of various sequencing artifacts, such as cell segmentation errors and imbalanced cell-type representations. Polyomino outperforms state-of-the-art methods by 10 to 1000 times in speed, and it is the only approach capable of integrating data sets containing millions of cells in a single run. As a result, Polyomino uncovers originally hidden gene expression patterns in brain sections and offers new insights into organogenesis and tumor microenvironments, all with exceptional efficiency and accuracy.

Usher syndrome (USH) is the most common form of syndromic deaf-blindness and is classified as a syndromic ciliopathy. It is characterized by sensorineural hearing loss, retinitis pigmentosa (RP), and, in some cases, vestibular dysfunction. USH is inherited in an autosomal recessive manner and presents in three distinct clinical subtypes: USH type 1 (USH1), type 2 (USH2), and type 3 (USH3). The estimated prevalence of USH ranges between 3 and 6 per 100,000 individuals worldwide. The syndrome is known for its clinical and genetic heterogeneity, with 13 genes currently identified as causative. At the molecular level, USH-related proteins are highly expressed in the periciliary region of retinal photoreceptors. This protein network is essential for maintaining the structural integrity of the photoreceptor cilium and facilitating intracellular transport between the inner and outer segments of photoreceptor cells. Disruption of these processes leads to progressive photoreceptor degeneration and vision loss, hallmark features of retinitis pigmentosa.

Deep learning-based segmentation models can accelerate the analysis of high-throughput microscopy data by automatically identifying and classifying cells in images. However, the datasets needed to train these models are typically assembled via laborious hand-annotation. This limits their scale and diversity, which in turn limits model performance. We present Cell-APP (Cellular Annotation and Perception Pipeline), a tool that automates the annotation of high-quality training data for transmitted-light (TL) cell segmentation. Cell-APP uses two inputs-paired TL and nuclear fluorescence images-and operates in two main steps. First, it extracts each cell's location from the nuclear fluorescence channel and provides these locations to promptable deep learning models to generate cell masks. Then, it classifies each cell as mitotic or nonmitotic based on nuclear features. Together, these masks and classifications form the basis for cell segmentation training data. By training vision-transformer-based models on Cell-APP-generated datasets, we demonstrate how Cell-APP enables the creation of both cell line-specific and multi-cell line segmentation models. Cell-APP thus empowers laboratories to tailor cell segmentation models to their needs and outlines a scalable path to creating general models for the research community.

Segmentation of white blood cells is critical for a wide range of applications. It aims to identify and isolate individual white blood cells from medical images, enabling accurate diagnosis and monitoring of diseases. In the last decade, many researchers have focused on this task using U-Net, one of the most used deep learning architectures. To further enhance segmentation accuracy and robustness, recent advances have explored the combination of U-Net with other techniques, such as attention mechanisms and aggregation techniques. However, a common challenge in white blood cell image segmentation is the similarity between the cells’ cytoplasm and other surrounding blood components, which often leads to inaccurate or incomplete segmentation due to difficulties in distinguishing low-contrast or subtle boundaries, leaving a significant gap for improvement. In this paper, we propose GAAD-U-Net, a novel architecture that integrates attention-augmented convolutions to better capture ambiguous boundaries and complex structures such as overlapping cells and low-contrast regions, followed by a gating mechanism to further suppress irrelevant feature information. These two key components are integrated in the Double U-Net base architecture. Our model achieves state-of-the-art performance on white blood cell benchmark datasets, with a 3.4% Dice score coefficient (DSC) improvement specifically on the SegPC-2021 dataset. The proposed model achieves superior performance as measured by mean the intersection over union (IoU) and DSC, with notably strong segmentation performance even for difficult images.

Semantic consistency-guided patch-wise relation graph reasoning scheme for lung cancer organoid segmentation in brightfield microscopy.

A Novel Weakly Supervised Immunohistochemical Cell Segmentation method via Counting Labels

Globally, breast cancer represents the leading cancer type, with millions of women impacted annually. The success of breast cancer treatment relies heavily on timely detection and precise tumor classification. The classification of breast cancer has gained considerable importance in Deep Learning (DL) and medical research with the development of medical imaging techniques, like histopathological imaging. Many existing DL schemes suffer from overfitting and endure difficulties in effectively mining the key features from high-resolution images with subtle variations. Hence, the Xception Convolutional Deep Maxout Network (Xcov-DMN) is developed to classify breast cancer. At the initial stage, the Mean-Shift Filter is applied to the input histopathological image. Following this, the White Blood Cell Network (WBC-Net) is employed for blood cell segmentation with the Balanced Cross-Entropy (BCE) and Focal Loss for ensuring precise segmentation. Next, Colored Histograms, shape features, Haralick Texture Features, and Complete Local Binary Pattern (CLBP) features are excerpted. Consequently, the developed Xcov-DMN is utilized to classify breast cancer. Xcov-DMN is the combination of the Deep Maxout Network (DMN), Fractional Calculus (FC), and Xception Convolutional Neural Network (XCovNet). Moreover, with learning data at 90%, the Xcov-DMN achieved the highest accuracy of 92.755%, True Negative Rate (TNR) of 91.977%, and True Positive Rate (TPR) of 94.765%.

The global burden of lung adenocarcinoma (LUAD) has been on the rise, making it among the leading contributor to cancer-related deaths. Long non-coding RNA (lncRNA) are implicated in the initiation and progression of LUAD. To date, the mechanism by which lncRNA participate in LUAD are not clearly characterized. Here, we investigated the role of the newly-discovered Lnc-PDZD7-3 in the development of LUAD. Results revealed downregulation of Lnc-PDZD7-3 in human normal lung tissues and upregulation in LUAD tissues from the TCGA (The Cancer Genome Atlas) databases. Excessive expression of Lnc-PDZD7-3 promotes occurrence of distant metastasis. Lnc-PDZD7-3 knockdown suppressed the proliferative and viability potential of cells, as well enhanced apoptosis and inhibited the migratory activity of LUAD cells. Notably, expression levels of MMP9, Vimentin, Twist, Fibronectin, and MMP2 in LUAD cells were downregulated markedly except for snail following Lnc-PDZD7-3 knockdown. Through rescue experiments, we confirmed that Lnc-PDZD7-3 enhanced LUAD development by activating FN1/fibronectin signaling. Meanwhile, we also identified that Lnc-PDZD7-3 was localized in cytoplasm and nucleus segments of LUAD cells by FISH technology. In summary, this study implicates Lnc-PDZD7-3 in the pathomechanisms of LUAD via the FN1/fibronectin signaling, suggesting it may be diagnostic biomarker and therapeutic targets of LUAD.

Recent advancements in spatial transcriptomics technologies have significantly enhanced resolution and throughput, underscoring an urgent need for systematic benchmarking. Here, we generate serial tissue sections from colon adenocarcinoma, hepatocellular carcinoma, and ovarian cancer samples for systematic evaluation. Using these uniformly processed samples, we generate spatial transcriptomics data across four high-throughput platforms with subcellular resolution: Stereo-seq v1.3, Visium HD FFPE, CosMx 6K, and Xenium 5K. To establish ground truth datasets, we profile proteins on tissue sections adjacent to all platforms using CODEX and perform single-cell RNA sequencing on the same samples. Leveraging manual nuclear segmentation and detailed annotations, we systematically assess each platform’s performance across capture sensitivity, specificity, diffusion control, cell segmentation, cell annotation, spatial clustering, and concordance with adjacent CODEX. The uniformly generated and processed multi-omics dataset could advance computational method development and biological discoveries. The dataset is accessible via SPATCH, a user-friendly web server for visualization and download.

Recent advancements in pathological imaging have facilitated single-cell and subcellular-level analysis based on high-resolution images for tumor subtyping, cytomorphological assessment, and infection detection. As high-resolution imaging is often limited by cost, super-resolution methods provide a practical alternative with only low-resolution data. However, existing methods generally suffer from artifacts, oversmoothing, and slow inference speed. In this study, we developed a hierarchal deep learning framework based on local pathological image patterns, named Hierarchical Local Image Patterns (HLIP), to achieve accurate, high-fidelity, and real-time super resolution with flexible magnifications. HLIP integrates semantic features with both pixel- and morphology-level features and reconstructs super-resolution images by the recognized local pathological image patterns. Benchmark analysis showed HLIP achieved the best performance and robustness on both internal and external test datasets. The generated super-resolution images contain abundant pathological details and maintain high fidelity. HLIP can be used for the enhancement of other models across multiple clinical scenarios, including gland segmentation, cell segmentation, Helicobacter pylori detection, and therapy response prediction. With its superior performance in pathology image super resolution, HLIP offers a versatile tool for image preprocessing in computer-aided systems, thereby supporting accurate diagnosis in clinical practice. © 2025 The Pathological Society of Great Britain and Ireland.

Red Blood Cell (RBC) deformability refers to the ability of RBCs to change their shape in response to external forces. This deformation plays a crucial role in altering the flow conditions within blood vessels, which in turn reduces the resistance to blood flow. RBC deformation is primarily caused by shear stress within the blood circulation. Various methods have been developed to measure RBC deformability, but these techniques often require specialized equipment, lengthy measurement times, and highly skilled personnel. So, to address these challenges, the novel technique, Gannet Puffer fish Optimization Algorithm with the Pyramid Network (GPOA_PyramidNet) is developed in this research for RBC deformability using microscopic images. Initially, the microscopic image from the Federated Research Data Repository (FRDR) is taken as an input. Then, the image enhancement is performed by the technique called Contract Limited Adaptive Histogram Equalization (CLAHE). Consequently, cell segmentation is carried out using the entropy-based Kapur methodology, in which GPOA selects the threshold values. The proposed GPOA is the fusion of the Pufferfish Optimization Algorithm (POA) and the Gannet Optimization Algorithm (GOA). Later, the features, like Gray-Level Co-occurrence Matrix (GLCM) and shape features are extracted. Then, the RBC deformability is achieved using the GPOA_PyramidNet, and they are classified as rigid RBCs and deformable RBCs. The proposed GPOA_PyramidNet is analyzed for its effectiveness in terms of metrics like accuracy, True Positive Rate (TPR), and True Negative Rate (TNR). It is found that the proposed method attained a maximum accuracy of 92.88%, TPR of 95.79%, and TNR of 93.90%, which is superior to the prevailing techniques. Moreover, the proposed GPOA_PyramidNet model achieves higher accuracy compared to several other methods, with improvements of 16.03% over Convolutional Neural Network (CNN), 11.84% over Convolution and Recurrent Neural Network (C-RNN), 7.52% over CNN-ALL, 2.13% over Variational Autoencoder (VAE), 1.52% over Image-based RBC deformability assessment via a shape-classification approach (IRIS), 1.08% over Light Weight Convolutional Neural Network (LWCNN), 0.96% over Region-based Convolutional Neural Network (RCNN), and 0.57% over ICAFF-MobileNetv2.

Accurate detection of mitosis is crucial in automated cell analysis, yet many existing methods depend heavily on deep learning models or complex detection techniques, which can be computationally intensive and error-prone, particularly when segmentation is incomplete. This study presents a novel unsupervised method for mitosis detection, leveraging the geometric properties of the Cassini oval to reduce computational costs and enhance robustness. Our approach integrates a newly developed deep learning model, MaxSigNet, for initial cell segmentation. We subsequently employ the Cassini oval in its single-ring mode to detect mother cells in the initial frame and switch to double-ring mode in subsequent frames to identify daughter cells and confirm mitosis events. The success of this method hinges on the presence of equal non-zero foci values in the mother cell and distinct non-zero foci values in the daughter cells, which indicate accurate mitosis detection. The method was evaluated across six datasets from four different cell lines, achieving perfect F1, Recall and Precision scores on four datasets, with scores of 96% and 85% on the remaining two. Comparative analysis demonstrated that our method outperformed similar approaches in F1 and Recall metrics. Additionally, the method showed substantial robustness to incomplete segmentation, with only a 20% average drop in F1 scores when tested with older segmentation methods like K-means, Felzenszwalb and Watershed. The proposed method offers a significant advancement in mitosis detection by leveraging the Cassini oval's properties, providing a reliable and efficient solution for automated cell analysis systems. This approach promises to enhance the accuracy and efficiency of cellular behaviour studies, with potential applications in various biomedical research fields.

Accurate cell segmentation is an essential step in the quantitative analysis of fluorescence microscopy images. Pre-trained deep learning models for automatic cell segmentation such as those offered by Cellpose perform well across a variety of biological datasets but may still introduce segmentation errors. While training custom models can improve accuracy, it often requires programming expertise and significant time, limiting the accessibility of automatic cell segmentation for many wet lab researchers. To address this gap, we developed "Toggle-Untoggle", a standalone desktop application that enables intuitive, code-free quality control of automated cell segmentation. Our tool integrates the latest Cellpose "cyto3" model, known for its robust performance across diverse cell types, while also supporting the "nuclei" model and user-specified custom models to provide flexibility for a range of segmentation tasks. Through a user-friendly graphical interface, users can interactively toggle individual segmented cells on or off, merge or draw cell masks, and export morphological features and cell outlines for downstream analysis. Here we demonstrate the utility of "Toggle-Untoggle" in enabling accurate, efficient single-cell analysis on real-world fluorescence microscopy data, with no coding skills required.

BACKGROUND: To date, there have been no studies that have investigated the role of mast cells in the development of sporadic medullary carcinoma of the thyroid. However, there is evidence that they are involved in the progression of epithelial malignancies of various localisations and as an independent predictor of long-term progression-free survival in neuroendocrine tumours of the pancreas. AIM: To show the possibilities of immunohistochemical detection of TC in the tumour microenvironment of sporadic MCRC. METHODS: The present study was conducted on histological sections of paraffin blocks of sporadic medullary thyroid carcinoma, with the use of immunohistochemical detection of mast cell tryptase. Subsequently, a CNN (convolutional neural network) model was trained for the segmentation of DAB-stained cells, followed by the calculation of the results. RESULTS: The results of the study revealed a number of possible clinically significant correlations, including: a correlation between the number of mast cells in the thyroid stroma and the age of patients; a correlation between the number of mast cells detected in the tumour at different stages according to the 8th edition of the TNM classification; and peculiarities of co-localisation of mast cells with other cells in the tumour environment. CONCLUSION: The study provides evidence for the presence of mast cells in the stroma of medullary carcinoma, and for quantitative differences in mast cell numbers at different node sizes. The active effect of TC on atypical cells of sporadic medullary thyroid carcinoma and other components of the tumour microenvironment, as detected in the study, is an important criterion for interpreting the biological effects of TC in relation to the tumour. Further analysis is required for the development of diagnostic algorithms and the improvement of prognosis objectivity.

Resolving spatial protein dynamics in native human epithelial tissues presents a significant technical challenge, particularly in inherently curved or unevenly mounted specimens. Here, we introduce an image processing pipeline for high-resolution, compartment-based analysis of protein localization, using the native three-dimensional architecture of the human anterior lens epithelium and capsule complex as a robust ex vivo proof-of-principle platform for precise cell segmentation and quantitative analysis. This platform integrates whole-mount immunostaining, 3D confocal imaging, computational tissue flattening, digital segmentation, and spatial regression to quantitatively map subcellular protein distributions at the tissue scale. To demonstrate the utility of this approach, we examined the spatial distribution of αB-crystallin (CRYAB), a stress-associated small heat shock protein, and βB2-crystallin (CRYBB2), a predominantly structural lens protein, in specimens obtained during cataract surgery. We observed a marked accumulation of CRYAB in epithelial cells at the capsule edge following both laser and manual capsulorhexis, indicating a localized stress response to surgical intervention. In contrast, CRYBB2 distribution remained unaffected. Furthermore, both proteins exhibited consistent cytoplasmic localization, while only CRYBB2 occasionally showed exclusive nuclear accumulation. This pipeline offers a scalable framework for quantitatively resolving protein localization in native epithelial architectures, using CRYAB and CRYBB2 as examples of how stress-associated changes can be spatially mapped in situ within the human lens.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-17896-0.

3DCellComposer - A versatile pipeline utilizing 2D cell segmentation methods for 3D cell segmentation.

One-shot cell segmentation via learning memory query: Towards universal solution without active tuning.

Abstract Mechanosensation plays a key role in many physiological processes. Despite its importance, high-content methods for studying mechanosignaling at the cell level remain challenging. This article presents an approach to assess mechanosignaling in adherent cells using isotropic mechanical stretch combined with real-time Ca2+ fluorescence imaging. This method integrates our IsoStretcher platform with image registration, cell segmentation, and single-cell and population-based Ca2+ signaling analysis.

Accurate detection and segmentation of fluorescent spots in microscopy cell images remain challenging. Traditional methods, based on centroid localization or pixel-wise semantic segmentation, often fail to delineate individual spot boundaries. This limitation significantly hinders the quantitative analysis of morphological heterogeneity and the interpretation of densely distributed subcellular signals. Here, we propose UPBAS-Net, a unified computational framework that integrates Fourier interpolation-based preprocessing with an enhanced YOLOv8 architecture incorporating an additional upsampling layer to improve shallow feature extraction and enable boundary-aware instance segmentation of fluorescent spots at subpixel resolution. It overcomes the limitations of traditional centroid localization and pixel-wise classification, enabling accurate delineation of spot boundaries. Experimental results show that UPBAS-Net achieves substantial improvements in spot localization accuracy, with F1-score gains up to 8.27% compared to the deepBlink model across multiple benchmark data sets. Furthermore, it demonstrates excellent scalability with the simultaneous segmentation of fluorescent spots and cellular boundaries, enabling integrated spatial correlation analysis at single-cell resolution. Additionally, we provide a user-friendly web-based analytical platform with containerized workflow management, enabling nonprogrammers to perform automated spot and cell segmentation using pretrained models. The platform is freely accessible at http://cellpropack.com/.