Nuclei Segmentation Research Articles

Comparison of Segmentation Analysis in Nucleus Detection with GLCM Features using Otsu and Polynomial Methods

Pap smear is a digital image generated from the recording of cervical cancer cell preparation. Images generated are susceptible to errors due to the relatively small cell sizes and overlapping cell nuclei. Therefore, accurate Pap smear image analysis is essential to obtain the right information. This research compares nucleus segmentation and detection using Grey Level Co-occurrence Matrix (GLCM) features in two methods: Otsu and Polynomial. The tested data consisted of 400 images sourced from RepoMedUNM, a publicly accessible repository containing 2,346 images. Both methods were compared and evaluated to obtain the most accurate features. The research results showed that the average distance of the Otsu method was 6.6457, which was superior to the Polynomial method with a value of 6.6215. Distance refers to the distance between the nucleus detected by the Otsu and the Polynomial method. Distance is an important measure to assess how closely the detection results align with the actual nucleus positions. It indicates that the Polynomial method produces nucleus detections that are on average closer to the actual nucleus positions compared to the Otsu method. Consequently, this research can serve as a reference for further studies in developing new methods to enhance the accuracy of identification.

A Cost-Effective Vital Sign Monitoring System Harnessing Smartwatch for Home Care Patients

Pap smear is a digital image generated from the recording of cervical cancer cell preparation. Images generated are susceptible to errors due to relatively small cell sizes and overlapping cell nuclei. Therefore, an accurate analysis of the Pap smear image is essential to obtain the right information. This research compares nucleus segmentation and detection using gray-level cooccurrence matrix (GLCM) features in two methods: Otsu and polynomial. The data tested consisted of 400 images sourced from RepoMedUNM, a publicly accessible repository containing 2,346 images. Both methods were compared and evaluated to obtain the most accurate characteristics. The research results showed that the average distance of the Otsu method was 6.6457, which was superior to the polynomial method with a value of 6.6215. Distance refers to the distance between the nucleus detected by the Otsu and the Polynomial method. Distance is an important measure to assess how closely the detection results align with the actual nucleus positions. It indicates that the polynomial method produces nucleus detections that are on average closer to the actual nucleus positions compared to the Otsu method. Consequently, this research can serve as a reference for future studies in developing new methods to enhance identification accuracy.

Histopathological Image Synthesis with Generative Adversarial Networks for Nuclei Segmentation

Histopathological Image Synthesis with Generative Adversarial Networks for Nuclei Segmentation

Erythrocyte Recovery in Oreochromis niloticus Fish Exposed to Urban Effluents.

Urban activities pollute aquatic ecosystems, and the integrity of organisms such as fish. The use of cytological techniques, such as the analysis of blood cellular integrity using the Micronucleus test, can help detect mutagenic damage as a result to urban effluents exposure. In this context, this study aimed to evaluate the frequency of micronucleus and other nuclear abnormalities in Oreochromis niloticus fish environmentally exposed to urban effluents in relation to their erythrocyte recovery capacity when exposed to clean water (30 and 45days). The results indicated high copper, dissolved iron, nickel, and thermotolerant coliform levels in the urban stream. There was no difference in the frequency of micronuclei. In contrast, cells with nuclear nuclei, binucleates, kidney-shaped nuclei, notched nuclei, lobed nuclei, and segmented nuclei decreased according to the time the fish were exposed to clean water. When exposed to clean water, we conclude that urban fish recover from genotoxic and cytotoxic damage.

Nuclei-Net: a multi-stage fusion model for nuclei segmentation in microscopy images

Nuclei-Net: a multi-stage fusion model for nuclei segmentation in microscopy images

Continual Nuclei Segmentation via Prototype-Wise Relation Distillation and Contrastive Learning.

Deep learning models have achieved remarkable success in multi-type nuclei segmentation. These models are mostly trained at once with the full annotation of all types of nuclei available, while lack the ability of continually learning new classes due to the problem of catastrophic forgetting. In this paper, we study the practical and important class-incremental continual learning problem, where the model is incrementally updated to new classes without accessing to previous data. We propose a novel continual nuclei segmentation method, to avoid forgetting knowledge of old classes and facilitate the learning of new classes, by achieving feature-level knowledge distillation with prototype-wise relation distillation and contrastive learning. Concretely, prototype-wise relation distillation imposes constraints on the inter-class relation similarity, encouraging the encoder to extract similar class distribution for old classes in the feature space. Prototype-wise contrastive learning with a hard sampling strategy enhances the intra-class compactness and inter-class separability of features, improving the performance on both old and new classes. Experiments on two multi-type nuclei segmentation benchmarks, i.e., MoNuSAC and CoNSeP, demonstrate the effectiveness of our method with superior performance over many competitive methods. Codes are available at https://github.com/zzw-szu/CoNuSeg.

Artifact Detection and Restoration in Histology Images With Stain-Style and Structural Preservation.

The artifacts in histology images may encumber the accurate interpretation of medical information and cause misdiagnosis. Accordingly, prepending manual quality control of artifacts considerably decreases the degree of automation. To close this gap, we propose a methodical pre-processing framework to detect and restore artifacts, which minimizes their impact on downstream AI diagnostic tasks. First, the artifact recognition network AR-Classifier first differentiates common artifacts from normal tissues, e.g., tissue folds, marking dye, tattoo pigment, spot, and out-of-focus, and also catalogs artifact patches by their restorability. Then, the succeeding artifact restoration network AR-CycleGAN performs de-artifact processing where stain styles and tissue structures can be maximally retained. We construct a benchmark for performance evaluation, curated from both clinically collected WSIs and public datasets of colorectal and breast cancer. The functional structures are compared with state-of-the-art methods, and also comprehensively evaluated by multiple metrics across multiple tasks, including artifact classification, artifact restoration, downstream diagnostic tasks of tumor classification and nuclei segmentation. The proposed system allows full automation of deep learning based histology image analysis without human intervention. Moreover, the structure-independent characteristic enables its processing with various artifact subtypes. The source code and data in this research are available at https://github.com/yunboer/AR-classifier-and-AR-CycleGAN.

Developing Topics.

Hippocampal and amygdala subfields variably affect cognitive impairment in neurodegenerative diseases. Subfields of these regions can be well segmented using modern neuroimaging tools but their role in neurodegenerative disease is under active investigation. In this study, we identified hippocampal and amygdala subregions predictive of cognitive performance and motor symptoms severity measured by MoCA (Montreal Cognitive Assessment) and MDS-UPDRS III, respectively, in patients with dementia with Lewy Bodies (DLB). We selected all participants with probable DLB (N = 48, mean age = 71±7 years, 15% female) enrolled in the Dementia with Lewy Bodies Consortium (DLBC) as of July 2022, with concurrent measures of 3D T1 MRI sequence, MoCA, and MDS-UPDRS III scores. We performed cortical reconstruction and volumetric segmentation of hippocampal subfields and nuclei of the amygdala using FreeSurfer (v 7.2). We used combat harmonization to account for site and scanner differences. We trained and applied a bootstrapped, bidirectional stepwise regression model of 29 predictor variables comprised of sub-fields and mean cortical thickness against MoCA and MDS-UPDRS III, respectively, with an 80-20 train-test split ratio, and 5000 repetitions, corrected for age and sex. Subfield segmentation is shown in Figure 1A. The best fitting model for MoCA included mean cortical thickness, parasubiculum, hippocampal and amygdala transition area, corticoamygdaloid transition area, and CA3 body (Figure 1B, adjusted R2 = 0.51). The best fitting model for MDS-UPDRS III included the cortical nucleus of the amygdala and CA1 body (Figure 1C, adjusted R2 = 0.22). This model was considered a poor fit. We considered MoCA for further analysis and closely predicted scores in our 20% partitioned test sample (Figure 2A, R2 = 0.38). We report model-based selection of hippocampal and amygdala subfields to predict MoCA scores in DLB. Atrophy in these regions has been associated with global cognitive deficit in mild cognitive impairment and Alzheimer disease cohorts. The model fit for MDS-UPDRS III scores was poor, providing evidence that these brain regions do not serve a role in motor control.

Information Added U-Net with Sharp Block for Nucleus Segmentation of Histopathology Images

Information Added U-Net with Sharp Block for Nucleus Segmentation of Histopathology Images

Chronic Myeloid Leukemia Presenting with Mixed Phenotype B/Myeloid Blast Phase: A Diagnostic Challenge

Abstract Introduction/Objective Chronic myeloid leukemia (CML) is typically identified in chronic phase, however about 5% of cases present in a more advanced phase. In blast phase (BP), the blasts are most commonly of myeloid lineage, less commonly lymphoid, and rarely mixed lineage cases are encountered. A mixed phenotype at presentation raises the possibility of a de novo mixed phenotype acute leukemia (MPAL) with BCR::ABL1 fusion and requires careful correlation with ancillary studies to confirm the diagnosis. Methods/Case Report The patient is a 60-year-old man who presented with leukocytosis, anemia, normal platelet count, and blasts identified on peripheral blood smear examination. A bone marrow biopsy demonstrated hypercellular marrow and a dimorphic population of large and small blasts, comprising nearly 70% of cellularity. Flow cytometry demonstrated distinct myeloid and B-lymphoid blast populations, both of which expressed CD13 and TdT. Interphase fluorescence in situ hybridization (FISH) showed the BCR::ABL1 fusion in more than 90% of nuclei, including segmented nuclei, and conventional cytogenetic studies showed t(9;22)(q34.1;q11.2) in 19 of 20 metaphases. Molecular studies demonstrated the minor BCR::ABL1 p190 transcript, as well as ASXL1 and dual RUNX1 mutations. The discrepant number of BCR::ABL1 positive cells by FISH and conventional cytogenetic studies compared to the proportion of blasts, and its presence in segmented nuclei, enabled a diagnosis of CML with mixed phenotype blast phase despite absence of features such as basophilia or significant granulocytic left shift. Results (if a Case Study enter NA) NA Conclusion We share this case of CML in mixed phenotype blast phase to highlight the importance of careful evaluation in distinguishing this rare presentation from de novo MPAL, particularly with BCR::ABL1 fusion. Features such as a disproportionate number of nuclei or metaphases carrying the t(9;22) translocation on FISH or conventional cytogenetic studies and the presence of BCR::ABL1 in mature granulocytes can be helpful when other morphologic features of CML are absent.

Enhancement of Human Neutrophil Bactericidal Activity through ROS Regulation

Antimicrobial agents are vital for treating infectious diseases, but their widespread use has led to a concerning rise in drug-resistant bacteria. The development and overuse of new antimicrobial agents have paradoxically given rise to even more drug-resistant strains. If current trends persist, drug-resistant infections may become the leading global cause of mortality by 2050. However, current measures such as restricting antimicrobial use and promoting appropriate usage have not provided a definitive solution. Therefore, exploring alternative treatment approaches is essential. In this study, our research was based on a novel strategy that shifts the focus from targeting specific bacterial species to leveraging the defense mechanisms of human neutrophils. Neutrophils play a critical role in innate immunity by engulfing bacteria and generating reactive oxygen species (ROS) in phagosomes. The primary ROS producer in neutrophils is the NADPH oxidase (NOX) complex centered on gp91 phox (NOX2), a membrane protein of the NOX family. However, excessive ROS are harmful, necessitating negative regulation. Neutrophils contain a negative regulator of reactive oxygen species (NRROS), which regulates ROS production by competing with p22 phox for binding to NOX2 when inflammation go into ending. NRROS knockout mice have shown increased ROS production and bactericidal activity during bacterial infection, resulting in significantly prolonged survival compared to wild-type mice. If this mechanism could be applied to human cells to enhance bactericidal activity, it would eliminate the need to target specific bacterial species and could revolutionize the treatment of infectious diseases. We used human induced pluripotent stem cells (iPSC) as a model to validate the NRROS knockout effect in human cells, overcoming challenges in isolating peripheral blood neutrophils due to their limited lifespan, poor in vitro culture tolerance, and genetic editing complexities. We successfully generated NRROS-heterozygous knockout (hetero-KO) and NRROS-homozygous knockout (homo-KO) iPSC clones using genome editing techniques. Additionally, using the PiggyBac system, we established doxycycline-inducible NRROS expression clones (homo-KO+Dox) based on homo-KO clones. These iPSC were differentiated into neutrophils and evaluated for function. The isolated neutrophil fraction exhibited characteristic segmented nuclei in all clones. We measured ROS production and examined changes induced by NRROS knockout after stimulation. Immediate superoxide production showed no difference between wild-type and NRROS-KO clones. However, after 2 hours of stimulation, the NRROS-KO clones exhibited significantly elevated hydrogen peroxide production, indicating increased total ROS production, possibly due to delayed NOX complex degradation. Furthermore, we assessed the bactericidal capacity of iPSC-derived neutrophils against multidrug-resistant bacteria, revealing significantly enhanced bactericidal activity in NRROS-KO clones compared to wild-type clones. In conclusion, NRROS knockout in human neutrophils enhances bactericidal capacity through heightened ROS production, offering promising prospects for innovative infectious disease treatments. If successfully harnessed, this innovative treatment modality could revolutionize infectious disease management, combating drug-resistant infections, and improving global health outcomes.

A Deep Learning-Based Pathomics Methodology for Quantifying and Characterizing Nucleated Cells in the Bone Marrow Microenvironment

Introduction: The overall cellularity of the bone marrow (BM) microenvironment is a WHO criterion for differentiating BCRABL- myeloproliferative neoplasms (MPNs) like polycythemia vera (PV) and myelofibrosis (MF) [1]. However, manual assessment by hematopathologists is subjective, not scalable, and only records a single global measure, disregarding BM regional differences. Leveraging pathology slide scanners and deep learning (DL) can address these limitations and reveal valuable insights into MPN subtypes and treatment effects. Despite its potential, computational pathology (Pathomics) in this area remains understudied [2-3]. Methods: After institutional review from Stony Brook University and Weill Cornell Medicine (WCM), we identified 54 patients (pts) with a manually confirmed PV diagnosis (dx) at WCM (17% newly dx'ed, 61% treated, 17% disease progression). The dataset includes hematoxylin & eosin (H&E) stained whole slide images (WSIs), labs, molecular studies, and pathology report elements (reference standard for biopsy core size and cellularity). Numeric cellularity measures the ratio of hematopoietic cells to fat cells and excludes inevaluable trabecular bone. Normal cellularity ranges from (100-pt age [y])±20% and PV is characterized by hypercellularity. We evaluated three DL algorithms for nuclear segmentation: 1) Our U-Net model trained on 5B nuclei from 10 cancer types [4], 2) HistoCartography (HC): a graph neural network trained on 235K nuclei (PanNuke/MoNuSac datasets) [5], and 3) CellPose (CP): a gradient-driven U-Net model trained on 1K nuclei of varied modalities [6]. A color/morphology-based segmenter identified fat cells and evaluable BM regions, excluding vessels/tissue tears. Cellular characteristics (shape, texture, color) were computed and clustered by k-nearest neighbors (kNN) and k-means to explore differences between PV stages. A tissue-based approach for cellularity estimation derived the proportion of non-fat tissue to evaluable regions. A cell-based approach derived the proportions of nucleated cell and fat cell areas. The study calculated mean differences between biopsy core size estimates, assessed correlation (r) and intra-class correlation (ICC) for numeric cellularity estimates, and Cohen's kappa (κ) for categorical cellularity estimates. Results: Measured and estimated biopsy core sizes differed by an acceptable margin (3±4 mm). Tissue-based cellularity estimator results: r=0.2, ICC=0.0, κ=0.2. Cell-based cellularity estimator result: our approach (r=0.7, ICC=0.6, κ=0.7), HC (r=0.7, ICC=0.5, κ=0.6), and CP (r=0.2, ICC=0.2, κ=0.7). Cell size clusters aligned with known distributions in PV, with the prominent Int1 class (5.1-9μm) likely containing erythrocytes and myeloid progenitors. Overall and cell-sized cellularity heatmaps visualize regional differences (Figure 1). K-means clustering separated new and progressing PV (Figure 2). Discussion: DL and classical image analysis were employed to segment nucleated cells, fat cells, and other evaluable regions of hematopoietic tissue. Extracting shape, texture, and color features from segmented nuclei enhances BM microenvironment characterization. This pilot study resulted in a practical and interpretable Pathomics pipeline and downstream analytics as a proof of concept for hypothesis-driven research. Future directions involve applying these insights to a large cohort of diagnosis WSIs from other MPN subtypes and correlating findings with other tools, like immune cell detection models.

Segmentation of nucleus based on dynamic convolution and deep features of stain distribution

AbstractAnalysis of the pathology image is important to diagnose cancer of lung, breast and stomach. Segmenting the nucleus is a key step for quantitative analysis, and has significance to the pathology researches and computer aided diagnosis systems. The inconsistency of colour, fuzzy boundary of nucleus and overlapping of cells are the universally acknowledged challenges. To solve these problems, the difference between the inside and outside of nucleus is enhanced by obtaining the distribution of the haematoxylin based on Lambert–Beer's law and the optical characteristics of stains. An inferior encoder, which is supervised by the inferior decoder, is proposed to extract the deep features of the distribution of stains. And these features are fed into the primary encoder to improve the accuracy of segmentation. To relieve the problem that some nuclei are segmented as background because the deep feature is inapparent, dynamic convolution is introduced into the encoders. The experiments show that the proposed model can segment the nucleus in the pathological images more precisely than the compared models. The Dice similarity coefficient (DSC) and panoptic quality (PQ) are 0.810 and 0.512, respectively.

Reversed domain adaptation for nuclei segmentation-based pathological image classification

Despite the fact that digital pathology has provided a new paradigm for modern medicine, the insufficiency of annotations for training remains a significant challenge. Due to the weak generalization abilities of deep-learning models, their performance is notably constrained in domains without sufficient annotations. Our research aims to enhance the model’s generalization ability through domain adaptation, increasing the prediction ability for the target domain data while only using the source domain labels for training. To further enhance classification performance, we introduce nuclei segmentation to provide the classifier with more diagnostically valuable nuclei information. In contrast to the general domain adaptation that generates source-like results in the target domain, we propose a reversed domain adaptation strategy that generates target-like results in the source domain, enabling the classification model to be more robust to inaccurate segmentation results. The proposed reversed unsupervised domain adaptation can effectively reduce the disparities in nuclei segmentation between the source and target domains without any target domain labels, leading to improved image classification performance in the target domain. The whole framework is designed in a unified manner so that the segmentation and classification modules can be trained jointly. Extensive experiments demonstrate that the proposed method significantly improves the classification performance in the target domain and outperforms existing general domain adaptation methods.

EPCO-34. IMAGE-BASED PHENOTYPIC HIGH-CONTENT CRISPR-CAS9 SCREEN FOR CHARACTERIZING CANCER DRIVERS IN PEDIATRIC HIGH-GRADE GLIOMA

Abstract The genomic landscape of pediatric high-grade glioma is characterized by a high level of aneuploidy but few focal amplifications or homozygous deletions. Cancer aneuploidies are the principal source of tumor heterogeneity driving tumor growth and therapy resistance. Here, we deployed an image-based phenotypic high-content CRISPR-Cas9 screen to identify aneuploidy-associated pediatric glioma driver genes. Through integrated computational analyses combining whole-genome sequencing datasets from 106 pediatric gliomas and 81 canine gliomas, we identified a set of 460 genes present in regions of aneuploidy in both diseases. We designed a custom guideRNA library to systemically knockout the 460 gene set using an arrayed CRISPR-Cas9 knockout screen in two pediatric high-grade glioma cell lines. By deploying 72 of 384-well plates on automated confocal Opera Phenix High-Content Screening System and dyes staining cell membrane, nucleus, and apoptotic cells, we established 5 million live-cell images across three timepoints (day 7, 8, 9 post-transduction). Each gene was randomized to 32 randomized wells across four 384-well plates to gather robust morphological profiles. Through image-based nucleus and cell segmentation workflow, we characterized 50 million single cells across 27,648 wells over three time points. Using semi-supervised machine learning methods, we derived 1,050 cell morphological features to characterize knockout phenotypes, e.g., viability, growth, and morphology (size, shape, texture of nucleus and cells) across cells. The top ranked genes were then linked to oncogenes and tumor suppressors based on pathway and ontology analysis as well as functional in vitro and in silico validation using Pediatric Cancer Dependency Map. We find convergence of the most impactful molecular abnormalities - based on their knockout phenotypes - on candidate signaling pathways for the development of new drugs and repurposing of existing drugs for pediatric high-grade glioma.

Histological classification of canine and feline lymphoma using a modular approach based on deep learning and advanced image processing

Histopathological examination of tissue samples is essential for identifying tumor malignancy and the diagnosis of different types of tumor. In the case of lymphoma classification, nuclear size of the neoplastic lymphocytes is one of the key features to differentiate the different subtypes. Based on the combination of artificial intelligence and advanced image processing, we provide a workflow for the classification of lymphoma with regards to their nuclear size (small, intermediate, and large). As the baseline for our workflow testing, we use a Unet++ model trained on histological images of canine lymphoma with individually labeled nuclei. As an alternative to the Unet++, we also used a publicly available pre-trained and unmodified instance segmentation model called Stardist to demonstrate that our modular classification workflow can be combined with different types of segmentation models if they can provide proper nuclei segmentation. Subsequent to nuclear segmentation, we optimize algorithmic parameters for accurate classification of nuclear size using a newly derived reference size and final image classification based on a pathologists-derived ground truth. Our image classification module achieves a classification accuracy of up to 92% on canine lymphoma data. Compared to the accuracy ranging from 66.67 to 84% achieved using measurements provided by three individual pathologists, our algorithm provides a higher accuracy level and reproducible results. Our workflow also demonstrates a high transferability to feline lymphoma, as shown by its accuracy of up to 84.21%, even though our workflow was not optimized for feline lymphoma images. By determining the nuclear size distribution in tumor areas, our workflow can assist pathologists in subtyping lymphoma based on the nuclei size and potentially improve reproducibility. Our proposed approach is modular and comprehensible, thus allowing adaptation for specific tasks and increasing the users’ trust in computer-assisted image classification.

Sharp dense U-Net: an enhanced dense U-Net architecture for nucleus segmentation

Sharp dense U-Net: an enhanced dense U-Net architecture for nucleus segmentation

Clinical Impact of the Spatial Organization of the Immune Tumor Microenvironment in Diffuse Large B-Cell Lymphoma

Introduction: Recent analyses of diffuse large B-cell lymphoma (DLBCL) have highlighted the clinical importance of immune tumor microenvironment (iTME), and based on the composition of the iTME, different DLBCL subtypes have been proposed (Kotlov et al. Cancer Discov. 2021, Steen et al. Cancer Cell. 2021). However, studies have mainly focused on the impact of different cell type proportions, whereas the clinical importance of their spatial organization has remained unclear. Materials and Methods: We used 12-plex immunohistochemistry panel to characterize B cells (CD20), T cells (CD3, CD4, CD8, FOXP3), macrophages (CD68, CD163), and immune checkpoint molecules (PD-1, PD-L1, CD96) from FFPE samples of 107 DLBCL patients treated with rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP)- like immunochemotherapy. We performed image processing using the Ilastik and CellProfiler softwares, and segmented nuclei with a pretrained deep learning segmentation model. Using histoCAT software, we quantified marker intensities for each cell, phenotyped the cells with the Phenograph algorithm, and, finally, performed a neighborhood analysis to recognize the cell types that neighbor or avoid each other. We correlated the findings with patient demographics and survival. Results: In total,we analyzed 739 825 single cells (median 7127 per sample; range 1640 - 12705) and discovered 16 different metaclusters, which included various T helper cell, cytotoxic T cell, regulatory T cell (Treg), M1 and M2 like macrophage, and B cell subgroups. Samples varied greatly in their immune cell composition, with a median proportion of B cells, T cells and macrophages being 48.0 %, 19.9 %, and 10.2 %, respectively. We divided the samples according to their iTME constitution using K means clustering. As expected, samples were split into non-inflamed (37 %) and inflamed iTME subgroups, the latter dominated by T cells (22 %) and M2 macrophages (40 %). However, there was no significant difference in survival between the subgroups. Neighborhood analysis revealed several interaction patterns, such as lymphoma cells favoring neighboring with other lymphoma/B cells. Interestingly, when T cells and, especially cytotoxic T cells, in the inflamed iTME neighbored with PD-L1/PD-1 negative B cells, the outcome was favorable (OS; p < 0.05; Figure 1A), independent of the IPI and cell-of-origin. In contrast, when B cells expressed PD-L1 or PD-1, there was no association with survival. Within the iTME, CD4+ T helper cells often neighbored with other CD4+ T cells, including Tregs, as well as cytotoxic T cells, and to a lesser extent macrophages. Cytotoxic T cells commonly neighbored with other cytotoxic T cells, but also with macrophages and T helper cells, whereas M1 and M2 like macrophages frequently neighbored with other M1 and M2 like macrophages, respectively, as well as T cells. When we clustered the cases according to how often M2 macrophages neighbored other immune cells, we identified a group of cases, where M2 macrophages accumulated around T cells. Notably, this iTME pattern translated to unfavorable outcome (OS; p < 0.05; Figure 1B). Conclusions: Our data reveal clinically significant interaction patterns between B and T cells, as well as between macrophages and T cells in the iTME of DLBCL.

RCKD: Response-Based Cross-Task Knowledge Distillation for Pathological Image Analysis.

We propose a novel transfer learning framework for pathological image analysis, the Response-based Cross-task Knowledge Distillation (RCKD), which improves the performance of the model by pretraining it on a large unlabeled dataset guided by a high-performance teacher model. RCKD first pretrains a student model to predict the nuclei segmentation results of the teacher model for unlabeled pathological images, and then fine-tunes the pretrained model for the downstream tasks, such as organ cancer sub-type classification and cancer region segmentation, using relatively small target datasets. Unlike conventional knowledge distillation, RCKD does not require that the target tasks of the teacher and student models be the same. Moreover, unlike conventional transfer learning, RCKD can transfer knowledge between models with different architectures. In addition, we propose a lightweight architecture, the Convolutional neural network with Spatial Attention by Transformers (CSAT), for processing high-resolution pathological images with limited memory and computation. CSAT exhibited a top-1 accuracy of 78.6% on ImageNet with only 3M parameters and 1.08 G multiply-accumulate (MAC) operations. When pretrained by RCKD, CSAT exhibited average classification and segmentation accuracies of 94.2% and 0.673 mIoU on six pathological image datasets, which is 4% and 0.043 mIoU higher than EfficientNet-B0, and 7.4% and 0.006 mIoU higher than ConvNextV2-Atto pretrained on ImageNet, respectively.

A Multi-Atlas Segmentation Algorithm with An Improved Sparse Representation on Brain MR Images

<p>Macaque brains are very close to human brains, so it’s an effective way to deepen the understanding of human brain functions by studying macaque brain structures. In order to segment subcortical nuclei of macaque brains more accurately, a multi-atlas segmentation algorithm based on an improved sparse representation has been designed in this paper. Firstly, a type of labeling information for atlas brain images is introduced when sparse patch-based representation is constructed, and then mutual information is improved by changing the calculation method of the information entropy, and it is used to measure the similarity between the target image and the atlas images. These two make the weights of the atlas more reasonable during fusion. Secondly, in order to fuse the segmentation results from two methods, nonlocal-patch-weighted method and the sparse representation method, a new similarity index based on a combination of dice coefficient and cosine distance is proposed. Finally, the experimental results show that this algorithm proposed in this paper has improved the accuracy of segmentation of hippocampus, striatum, claustrum and other nuclei, and it has better robustness.</p> <p> </p>