Imaging-based spatial transcriptomics enables high-resolution spatial mapping of RNA species. A key challenge in imaging-based spatial transcriptomics is accurate cell segmentation to assign each RNA molecule to the right cell. Here, we present RNA2seg, a novel segmentation algorithm trained on over 4 million cells from MERFISH and CosMx datasets across seven organs using a teacher-student training scheme. RNA2seg integrates RNA point clouds and all available membrane and nuclear stainings. Validation on manually annotated data shows superior performance including in zero-shot and few-shot settings.

Atypical mitotic figures (AMFs) indicate abnormal cell division associated with poor prognosis. Their detection remains difficult due to low prevalence, subtle morphology, and inter-observer variability. The MItosis DOmain Generalization (MIDOG) 2025 challenge introduces a benchmark for AMF classification across multiple domains. In this work, we fine-tuned the recently published DINOv3-H+ vision transformer, pretrained on natural images, using low-rank adaptation (LoRA), training only 1.3M parameters. We combine this with extensive augmentation and a domain-weighted Focal Loss to better handle the strong domain heterogeneity in the dataset. Despite the large shift between natural images and histopathology, our fine-tuned DINOv3 transfers effectively, reaching first place on the final test set. These results highlight the advantages of DINOv3 pretraining and underline the efficiency and robustness of our fine-tuning strategy, yielding state-of-the-art results for the atypical mitosis classification challenge in MIDOG 2025. Our code is publicly available on GitHub ^1 . ^1 ( https://github.com/Sanofi-Public/EFTD-midog-amf )

Agreement between two pathologists

CosMx protein panel showing stainings in MGC

Immunohistochemistry for representative Visium markers

Automated mitosis detection is a well-established task in computational pathology. While previous benchmarks focused on scanner-induced domain shift, clinical "real-world" application requires models to be robust across the vast variance to be expected in the histological landscape. The MItosis DOmain Generalization (MIDOG) 2025 challenge was designed to evaluate algorithmic performance across unprecedented biological and contextual diversity. We curated a test dataset of 365 cases, encompassing 12 distinct human, canine and feline tumor types, digitized across multiple scanning platforms. Moving beyond hand-selected hotspots, the challenge required detection also in random tissue areas (representative of the whole slide detection situation) and challenging areas (areas rich in hard negatives). In the second track, we introduced the classification of atypical mitotic figures (AMFs). There were 18 teams submitting to the detection track, with F1 scores ranging up to 0.740. In the AMF detection track, we had 21 submissions with balanced accuracy values up to 0.908. Our analysis reveals that while most models perform reliably in traditional hotspots, significant performance degradation occurs in challenging ROIs, where false positive rates tripled. Furthermore, performance varied significantly across the 12 tumor types, highlighting "blind spots" in current state-of-the-art architectures when encountering rare or highly pleomorphic malignancies. Moreover, we evaluated the effectiveness of ensembling and found a mean increases of 1.5 and 1.3 percentage points in F1 score and balanced accuracy, respectively. In contrast, TTA showed no relevant improvement. MIDOG 2025 demonstrates that "in the wild" mitosis detection remains a significant hurdle. The transition from hotspot-only evaluation to a multi-contextual framework provides a more realistic proxy for clinical reliability.

Abstract Understanding the role of gene expression in cellular function and tissue organization requires spatial and quantitative detection of individual RNA molecules. Yet, the widespread adoption of automated single-molecule fluorescence in situ hybridization (smFISH) has been limited by the cost of equipment and the complexity of experimental procedures. We present autoFISH, an affordable, user-friendly platform that removes these barriers through open-source hardware components, accessible control software, and integrated analysis tools. The system demonstrates broad applicability by enabling both conventional and signal-amplified smFISH protocols and incorporates an optimized tissue-clearing method that preserves nuclear structures. Testing across multiple cell types and tissue preparations validates the system’s reliability and reproducibility, offering a practical solution for scaling spatial transcriptomics research and advancing discoveries in cellular and developmental biology, while significantly reducing costs and the technical expertise required.

Understanding the spatial organization of individual cell types within tissue and how this organization is disrupted in disease, is a central question in biology and medicine. Hematoxylin and eosin-stained slides are widely available and provide detailed morphological context, while spatial gene expression profiling offers complementary molecular insights, though it remains costly and limited in accessibility. Predicting gene expression directly from histological images is therefore an attractive goal. However, existing approaches typically rely on small image patches, limiting resolution and the ability to capture fine-grained morphological variation. Here, we introduce a deep learning approach that predicts single-cell gene expression from morphology, matching patch-based methods on spot level prediction tasks. The model recovers biologically meaningful expression patterns across two cancer datasets and distinguishes fine cell populations. This approach enables molecular-level interpretation of standard histological slides at scale, offering new opportunities to study tissue organization and cellular diversity in health and disease.

Histopathological analysis is a cornerstone of cancer diagnosis, with Hematoxylin and Eosin (H&E) staining routinely acquired for every patient to visualize cell morphology and tissue architecture. On the other hand, multiplex immunofluorescence (mIF) enables more precise cell type identification via proteomic markers, but has yet to achieve widespread clinical adoption due to cost and logistical constraints. To bridge this gap, we introduce MIPHEI (Multiplex Immunofluorescence Prediction from H&E Images), a U-Net-inspired architecture that leverages a ViT pathology foundation model as an encoder to predict mIF signals from H&E images using rich pretrained representations. MIPHEI targets a comprehensive panel of markers spanning nuclear content, immune lineages (T cells, B cells, myeloid), epithelium, stroma, vasculature, and proliferation. We train our model using the publicly available OrionCRC dataset of restained H&E and mIF images from colorectal cancer tissue, and validate it on five independent datasets: HEMIT, PathoCell, IMMUcan, Lizard and PanNuke. On OrionCRC test set, MIPHEI achieves accurate cell-type classification from H&E alone, with F1 scores of 0.93 for Pan-CK, 0.83 for α-SMA, 0.68 for CD3e, 0.36 for CD20, and 0.28 for CD68, substantially outperforming both a state-of-the-art baseline and a random classifier for most markers. Our results indicate that, for some molecular markers, our model captures the complex relationships between nuclear morphologies in their tissue context, as visible in H&E images and molecular markers defining specific cell types. MIPHEI offers a promising step toward enabling cell-type-aware analysis of large-scale H&E datasets, in view of uncovering relationships between spatial cellular organization and patient outcomes.

Recent advances in pathology foundation models have enabled accurate prediction of spatial transcriptomics (ST) from routine H E images. However, existing explainability methods for vision transformer (ViT)-based models are largely limited to local heatmaps and do not reveal how morphological concepts contribute to ST predictions. Here, we introduce an explainable framework that combines relevance propagation and concept discovery to link transcriptional programs to tissue morphology. We developed a ViT-based framework for virtual ST from H E images that combines ViT-aware layer-wise relevance propagation with relaxed archetypal TopK sparse autoencoder-based concept discovery. This approach provides both local explanations and global insights into the morphological patterns associated with transcriptional programs. We applied the framework to colorectal cancer ST data from the HEST-1k cohort and evaluated its generalizability in TCGA COAD. Our architecture accurately predicts clinically relevant ST signatures and accompanying molecular phenotypes. Measured and predicted gene expression profiles reveal substantial spatial heterogeneity of the colorectal cancer subtypes iCMS2 and iCMS3 across a large number of samples. Spatially resolved and aggregated iCMS classification achieve weighted F1 scores of 0.872 and 0.819 (0.770 in TCGA COAD), respectively, and both stratify patient outcome. Beyond prediction, our framework establishes a relevance-based concept atlas linking molecular phenotypes to histopathological representations. Comparison of activation- with relevance-derived concepts demonstrates that relevances provide a more direct link between tissue morphology and downstream predictions. We establish a general strategy for concept-based explanation of spatial prediction, and our framework is readily applicable to a broad range of ViT-based pathology models.

CosMx 64-plex human immuno-oncology protein panel

Quantification of TILs, CD68-positive mononuclear macrophages and tumor grade

Representative video of an isolated MGC in oral cavity SCC from GR cohort

This table provides the DEG for the seven main clusters of the Visium experiment

Performances of the deep-learning cell detection model in TCGA cohorts (oral cavity, uterine cervix and larynx SCC) and GR cohorts (oral cavity SCC)

Clinical Tables and Statistical Analyses

Multiplex Immunofluorescence (mIF) provides detailed information on the abundance and spatial localization of clinically relevant proteins in tissue, yet its acquisition is costly and not widely accessible. Hematoxylin and eosin (H&E) staining, the pathological gold standard, is inexpensive and ubiquitous but lacks molecular specificity. Recent deep learning models can predict mIF signals from H&E, enabling in silico virtual staining, but standardized, publicly reproducible benchmarks are lacking, making it difficult to compare published methods. We propose a multi-level benchmark for $\mathrm{H} \& \mathrm{E} \rightarrow$ mIF prediction using public datasets. Representative models are trained on a shared dataset and evaluated across colorectal cancer cohorts. Performance is assessed at the pixel level for mIF reconstruction, at the cell level for cell-type classification, and with respect to cross-dataset generalization and computational cost. Our protocol provides a unified, reproducible framework to compare virtual-staining models in terms of both quantitative accuracy and biological relevance. Code and pretrained models are available on GitHub11https://github.com/Sanofi-Public/MIPHEI-ViT.

Association between keratin, MGC and clinical outcome

Abstract How gene expression patterns change spatially as the embryo transitions from simple to complex structures remains a major developmental biology question. Recently developed imaging-based spatial transcriptomics (ST) enable mapping expression of multiple gene at a single-cell resolution. Although Xenopus is a key model in embryology there is no established ST pipeline, and commercially available techniques face many challenges (sample preparation, probe design, cell segmentation). Furthermore, the highly diverse cell shapes and sizes across developmental stages and between different tissues represent major hurdles to accurately defining cells. Here, we describe an optimized workflow for ST in blastula-to-tailbud-stage frog embryos using Merscope, commercial MERFISH (Multiplexed Error-Robust Fluorescence In Situ Hybridization) originally designed for standard mammalian tissues. With stringent quality control and tailored computational pipelines, we optimize this technology for robust, semi-quantitative profiling of spatial transcriptomic landscapes in non-mammalian embryos. Reliable tissue preservation and cell-segmentation enable high-resolution mapping of gene expression during the development of a complex multi-tissue organization. This versatile strategy applies broadly to various dynamic systems, from embryos of various model organisms to complex and heterogeneous organs in mammals. Summary statement This Single-cell Spatial Transcriptomics pipeline and reference atlas in Xenopus - a model organism in embryology - overcome technical challenges and resolve dynamic changes in patterning during development.