single-cell morphology classification

Existing cell morphology quantification models are typically trained on single microscopy modalities, exhibiting poor generalizability across technical platforms (e.g., varying channel counts) and biological contexts. Method: We introduce CHAMMI-75—the first open, multi-channel pretraining dataset comprising 75 heterogeneous biological studies—enabling systematic integration of cross-platform, multi-channel, and multimodal microscopic images. We propose a channel-agnostic pretraining paradigm incorporating heterogeneous image registration, channel-wise normalization, and modality-aware multi-scale augmentation. Contribution/Results: Our approach significantly improves model robustness and generalization across diverse downstream tasks, empirically validating the critical role of modality diversity in cell morphology modeling. This work establishes a novel paradigm and provides essential resources for developing reusable, scalable foundation models in bioimage analysis.

To address the trade-off between classification accuracy and computational efficiency in single-cell RNA sequencing (scRNA-seq) data, this paper proposes EnProCell—a unified low-dimensional projection framework. EnProCell jointly optimizes principal component analysis (PCA) for expression variance preservation and multi-discriminant analysis (MDA) for inter-class separability, yielding a discriminative yet information-rich low-dimensional embedding space; a lightweight deep neural network is then trained in this space for efficient cell-type classification. Its key innovation is the first zero-shot transfer prediction for unlabeled query samples—requiring neither fine-tuning nor additional annotations. Evaluated on four major scRNA-seq platforms, EnProCell achieves state-of-the-art performance with significantly reduced computational overhead: 98.91% accuracy (F1 = 98.64%) on reference data and 99.52% accuracy (F1 = 99.07%) on unseen query data.

Manual annotation of cells and structures in whole-slide tissue images is prohibitively expensive. This work proposes an end-to-end “cluster-then-annotate” pipeline that dramatically reduces labeling effort by shifting the annotation target from individual structures to representative clusters. The method leverages Cellpose-SAM for structural segmentation, ResNet-50 to extract morphological embeddings, and UMAP followed by DBSCAN for dimensionality reduction and clustering. Annotators label only one exemplar per cluster, enabling comprehensive coverage of the entire image. Evaluated across 13 tissue types and 3,696 annotated structures, the approach achieves a weighted cluster-to-label alignment accuracy of 96.8%, with perfect consistency in seven tissue categories. This framework demonstrates the first successful generalization of morphology-based clustering across species and diverse tissue types, substantially enhancing annotation efficiency.

This study addresses the reliance on manual annotations and stage separation in cell segmentation and phenotypic classification for microscopy images by proposing an unsupervised generative routing pyramid framework. Through a coarse-to-fine reconstruction mechanism that links pixels to sparse latent sources, this approach unifies instance segmentation with phenotypic representation learning, thereby eliminating dependency on labeled data. Experimental results demonstrate that the model achieves competitive segmentation performance across diverse cell morphologies and imaging modalities while effectively modeling cellular phenotypes under perturbation conditions. Consequently, this work establishes a novel paradigm for unsupervised microscopy image analysis, offering a robust solution for integrated cell parsing without supervision.

Cell morphological heterogeneity severely impedes mechanistic studies of cellular regulation, yet existing approaches lack quantitative, comparable morphological representations and rely on subjective manual annotation. To address this, we propose a topological data analysis (TDA)-based single-cell morphological modeling framework: radial distance functions are constructed from cell contours and nuclear positions, yielding persistence diagrams as computable, comparable morphological signatures. A population-level distance matrix is derived via Wasserstein distances between these diagrams; combined with multi-linkage hierarchical clustering and a novel purity score, the framework enables unsupervised subpopulation identification and quantitative assessment of population homogeneity. This work represents the first systematic application of persistence diagrams to single-cell morphology characterization. Validated on human mesenchymal stem cells, it successfully resolves biologically meaningful subpopulations, demonstrating superior sensitivity to heterogeneity, robustness against noise, and significantly outperforming both manual annotation and conventional shape-parameter statistics.

本文提出一种端到端框架，通过实例感知图建模解决细胞检测和分类问题，动态构建图结构并融合视觉与关系证据，优于现有方法。

This work addresses the challenge that in single-cell perturbation data, cell populations under different perturbations exhibit substantial overlap, rendering conventional single-cell classification accuracy an unreliable metric of model performance. To overcome this limitation, the authors propose the Classifier Discrimination Score (CDS), which constructs a perturbation-level profile by aggregating classifier output probability distributions across entire cell populations and replaces single-cell predictions with population-level ranking. Remarkably, CDS recovers near-perfect perturbation identification from weak classifiers without requiring retraining. The method is compatible with diverse architectures—including linear models, MLPs, and Transformers—and demonstrates significant gains in identification accuracy on the Tahoe-100M and Virtual Cell Challenge datasets, with particularly pronounced advantages in low-cell-count regimes.

This work addresses the limitation of existing self-supervised learning methods in single-cell microscopy imaging, which often neglect hierarchical structures and consequently obscure fine-grained morphological features under coarse-grained factors. To overcome this, the authors propose the first self-supervised learning framework that explicitly models hierarchical organization by integrating segmentation-based teacher distillation with a hierarchical contrastive loss derived from HDBSCAN. This approach effectively disentangles coarse- and fine-grained features, yielding biologically meaningful cellular subclusters. Evaluated across 20 datasets encompassing 2.3 million cells, the method achieves an average top-K accuracy improvement of 2.8%, a 6.3% gain in top-9 retrieval performance at the deepest hierarchy level, and a 7.8% increase in F1-score for drug perturbation classification.

研究使用多标签分类方法解决血液细胞及其聚集体在微流控通道中的识别问题，相比传统的多类分类，该方法能更好地识别未出现在训练数据中的细胞聚集体。

This study addresses the challenge of fine-grained classification within the ambiguous “Other” phenotype category in zebrafish embryo images by proposing a two-stage hierarchical ensemble approach. In the first stage, a four-class model identifies major phenotypes; in the second stage, three specialized ensemble architectures further refine the “Other” class, with Setup 2—incorporating a multi-label classifier—demonstrating superior performance and better class balance. Experiments conducted using ResNet18, Vision Transformer (ViT), and ConvNeXt backbones reveal that ConvNeXt substantially enhances feature representation and consistently achieves the best results across all configurations, thereby validating the efficacy and advancement of the proposed hierarchical ensemble strategy.