Search Results (407)

Background/Objectives: Multiple instance learning (MIL) is widely used for automated classification of epithelial ovarian cancer subtypes from whole slide images (WSIs), but the relative contributions of feature extractor, aggregation strategy, and classification framework (flat vs. hierarchical) choices remain unclear under severe class imbalance. Methods: We evaluated 36 configurations on 510 WSIs from the UBC-OCEAN dataset using stratified five-fold cross-validation, comparing three pathology foundation models (Phikon-v2, CTransPath, UNI), six aggregators (mean/max pooling, ABMIL, CLAM-SB, DSMIL, DTP-TransMIL), and two classification strategies. Pathologist-annotated WSIs assessed attention map interpretability. Results: Feature extractor selection contributed substantially more variance than aggregator choice. Cascade balanced accuracy ranged from 0.538 (Phikon-v2) to 0.925 (UNI); CTransPath (~32 K pretraining WSIs) reached 0.870, exceeding Phikon-v2 (~58 K WSIs) and approaching UNI (~100 K+ WSIs), indicating that pretraining objective and architecture contribute as substantially as scale. The hierarchical cascade consistently improved high-grade serous carcinoma (HGSC) recall across all six evaluated configurations (+0.073 to +0.530), detecting 206 of 217 cases (0.949) with UNI max pooling. Quantitative spatial alignment analysis confirmed that both stronger feature extractors—CTransPath and UNI—generated significantly more spatially structured attention distributions than Phikon-v2 (paired Wilcoxon, p = 0.008 and p = 0.032, respectively). Conclusions: Feature extractor choice contributed more variance than aggregator selection, with the largest gap between Phikon-v2 and stronger extractors. Hierarchical cascades consistently improved HGSC recall across all configurations. Full article

Foundation models are reshaping computational pathology by enabling scalable task-agnostic representations of histopathological whole-slide images (WSIs). Unlike earlier task-specific deep learning systems, pathology foundation models (PFMs) leverage massive whole-slide image repositories and self-supervised Vision Transformer architectures to achieve broad generalization and few-shot adaptability. Their evolution reflects a shift from weakly supervised approaches such as Clustering-Constrained Attention Multiple Instance Learning (CLAM) and hierarchical architectures such as Hierarchical Image Pyramid Transformer (HIPT) to large-scale efforts including foundation models, UNI, Virchow, Phikon, CONtrastive learning from Captions for Histopathology (CONCH), GigaPath, H-Optimus, Transformer-Based Pathology Image and Text Alignment Network (TITAN), and the Mayo Clinic Atlas. These models demonstrate impressive performance across diagnostic and prognostic benchmarks while also opening pathways for multimodal integration with genomics and clinical data. Yet significant barriers remain including inconsistent generalization across institutions, interpretability lagging behind clinical needs, and slow integration into routine laboratory workflows. Certain domains of anatomic pathology such as cytopathology, transplant pathology, frozen sections, and rare tumor subtypes remain particularly resistant to current models. Here, we review the development of PFMs, critically evaluate their strengths and limitations, and outline priorities for their safe and effective clinical translation. We argue that the next phase of PFM development will depend on rigorous benchmarking, pathologist-in-the-loop deployment, and multimodal fusion ensuring these models evolve from research tools into clinically robust systems. Full article

Background/Objectives: Ovarian epithelial cancers (EOCs) comprise heterogeneous subtypes with distinct clinical outcomes, making accurate histological subtyping essential for prognosis and treatment planning. Although deep learning using digitized hematoxylin and eosin (H&E) whole-slide images (WSIs) is now widely used, its application to ovarian cancer diagnosis remains limited. Methods: In this multicenter study, we analyzed 319 H&E-stained slides from 152 patients with surgically resected EOC. An attention-based multiple instance learning (MIL) framework built on a pathology-specific foundation model (UNI) was used. WSIs were divided into 512 × 512-pixel patches at 40× magnification, and slide-level classification were generated through attention-based aggregation of patch-level feature, followed by patient-level prediction. External validation was performed specifically on the high-grade serous carcinoma (HGSC) cases from The Cancer Genome Atlas (TCGA) dataset. Results: The model achieved strong performance, with slide-level and patient-level accuracies of 0.918 and 0.900, respectively, on the test set. In five-fold cross-validation, the mean slide-level AUC was 0.990 (95% CI: 0.983–0.997), and the patient-level AUC was 0.993 (95% CI: 0.989–0.996), indicating consistent results. External validation on TCGA HGSC cases showed robust generalizability, with slide-level and patient-level accuracies of 0.794 and 0.898. F1-scores ranged from 0.832 to 1.000 at the slide-level and from 0.831 to 0.966 at the patient-level, with particularly strong performance for HGSC and clear-cell carcinoma. Conclusions: These findings demonstrate the feasibility of deep learning-based models for histological subtyping of EOC using H&E-stained WSIs. This approach may help pathologists achieve more accurate and consistent histological diagnoses of EOC. Full article

Synaptogenesis-related neuron–glioma interactions are increasingly recognized in glioma, yet it remains unclear whether routine H&E morphology can capture these programs and improve prognostic stratification. We integrated H&E whole-slide images, transcriptomes, and clinical data from 434 TCGA gliomas. Deep learning and quantitative pathology yielded an integrated histomorphologic feature set of 2678 features. Synaptogenesis-related activity was quantified using ssGSEA for ninety-eight synaptogenesis-related genes. In the training cohort, Spearman analysis identified 149 correlated histomorphologic features, which were refined to thirty-five by elastic net regularization. Seventeen prognostic candidates were entered into the MIME1 framework, and the most parsimonious model, Enet[0.1], retained fourteen non-zero-coefficient features to define the synaptogenesis-associated histomorphologic signature and construct the pathology-derived risk score (PRS). Multi-omic analyses, Human Protein Atlas validation, and single-nucleus RNA-seq were used to investigate the hub gene and its cellular context. PRS robustly stratified survival in both training and validation cohorts and remained an independent prognostic factor after adjustment for age and 2021 WHO CNS grade. High-risk tumors showed increased stromal and immune scores and enrichment of immune, adhesion, and phagosome-related pathways. EFNB2 emerged as the hub gene and was enriched in glioblastoma, and EFNB2-positive malignant cells displayed prominent communication with neurons, including EFNB2-EPHB1 signaling. Exploratory re-analysis of the myeloid compartment further showed that glioblastoma was enriched for suppressive TAM-like states relative to astrocytoma grade 2, supporting a shift toward a more tumor-associated and potentially immunosuppressive microenvironment. Routine H&E histomorphology can capture synaptogenesis-related molecular programs in glioma. The resulting PRS provides clinically relevant prognostic stratification, while EFNB2-positive malignant cells may represent a candidate hub for neuron–tumor communication within a remodeled tumor ecosystem. Full article

Background/Objectives: The selection of neoadjuvant therapy for patients with non-metastatic pancreatic adenocarcinoma remains challenging. Methods: We performed a single-institution, retrospective analysis of 79 patients who underwent resection of their pancreatic adenocarcinoma after receiving neoadjuvant therapy. Clinical and pathologic data were collected. Tumor fibrosis was quantified using Masson’s trichrome staining, tumor-infiltrating lymphocytes (TIL) were evaluated by an AI-based analysis of whole-slide H&E images, and immune cell populations were quantified by multiplex immunohistochemistry. Correlation analyses were performed between neoadjuvant treatment regimen, tumor regression, immune phenotypes, and survival. Results: All patients received chemotherapy, 77% FOLFIRINOX and 23% Gemcitabine/nab-paclitaxel (Abraxane). Eighteen percent of patients went on to receive radiation. Tumor regression grade (TRG) correlated with the neoadjuvant regimen. A reduction in tumor markers and the baseline neutrophil-to-lymphocyte ratio (NLR) correlated with overall survival. Among patients with an NLR > 3.3, FOLFIRINOX conferred a survival benefit over Gemcitabine/nab-paclitaxel, and radiation trended towards improved survival. Radiation was associated with increased fibrosis and reduced infiltration of CD8⁺ and regulatory T cells (Tregs). Increased Tregs and PDL1⁺ stromal cells were associated with poor response to neoadjuvant therapy, and NLR > 3.3 correlated with increased Treg infiltration. Conclusions: Our data suggest that patients with a high baseline NLR may benefit from intensified neoadjuvant therapy with FOLFIRINOX and radiation. Combination immunotherapy targeting Tregs and the PD1/PDL1 axis may further improve outcomes. Full article

Spatial transcriptomics (ST) has emerged as an essential technology for interpreting the molecular profiles underlying pathological tissue morphology. Most existing ST analyses are limited to 2D sections, which ignore the complex structural and molecular heterogeneity of biological tissues in 3D space and may cause diagnostic oversights. Since acquiring complete 3D ST volumes is resource-intensive, recent 3D imputation paradigms provide a cost-effective alternative by integrating 3D whole-slide images (WSIs) with sparse 2D ST references (e.g., a single slide). Despite this methodological advancement, effectively modeling complex cross-layer spatial dependencies remains challenging. Current mainstream solutions predominantly adopt standard Transformers for cross-scale feature aggregation, which may bring computational overhead and higher overfitting risk while having limited explicit mechanisms for hierarchical anatomical guidance. To address these limitations, we propose a Hierarchical MambaOut-based Spatial Imputation Graph Network (HM-ASIGN) for anatomy-aware 3D spatial transcriptomics imputation. Our architecture leverages MambaOut’s dynamic gated 1D convolutions as a parameter-efficient alternative to dense global self-attention. This design captures the depth-wise evolution of pathological features while reducing over-parameterization. Inspired by the macro-to-micro diagnostic reasoning of clinical pathologists, HM-ASIGN introduces a multi-scale recursive guidance mechanism. It constructs a top-down information flow by extracting global anatomical priors at macroscopic scales and injecting them as contextual anchors into regional and spot-level features in a cascaded manner. This helps ensure that fine-grained molecular predictions are properly constrained by global morphological structures. Evaluation experiments on multiple public breast cancer datasets demonstrate that HM-ASIGN achieves competitive reference-level performance against existing baselines, reaching a Pearson Correlation Coefficient (PCC) of 0.772. Specifically, when evaluated against the foundational ASIGN framework, it improves predictive accuracy while reducing the total parameter count by approximately 33.3% and improving inference throughput. Our results suggest that HM-ASIGN provides a computationally efficient approach for 3D spatial molecular mapping. Full article

Background: Accurate prognostic stratification in hepatocellular carcinoma (HCC) remains challenging due to substantial tumor heterogeneity. Deep learning (DL)-based approaches, particularly convolutional neural networks (CNNs), have been widely applied for prognostic prediction using histopathological images. Recently, pathology foundation models combined with multiple-instance learning (MIL) have emerged as a promising alternative. In this study, we aimed to provide a comparative evaluation of conventional CNN-based and foundation model-based approaches for HCC prognosis prediction, with a focus on tissue selection strategies and cross-dataset generalizability. Methods: We compared a patch-level CNN approach with a foundation model-based MIL approach for overall survival (OS) prediction from hematoxylin and eosin-stained whole-slide images. A total of 256 patients from Seoul St. Mary’s Hospital (SSMH) and 334 patients from the TCGA dataset were included. Models were trained using either all tissue regions or tumor-only regions under five-fold cross-validation. Model performance was evaluated using the concordance index (C-index), Kaplan–Meier analysis, and time-dependent receiver operating characteristic analysis. Cross-dataset validation and combined-dataset training assessed generalizability. Results: In the SSMH dataset, the CNN model performed better with tumor-only regions (C-index 0.8308) than with all tissue regions (0.7498). In contrast, the foundation model-based MIL approach showed stable performance regardless of input regions (C-indices: 0.8701 and 0.8752). Similar stability was observed in the TCGA dataset (0.7744 and 0.7722). Cross-dataset validation showed reduced performance, indicating limited generalizability. Combining datasets did not lead to performance improvement. Subgroup analyses showed prognostic information beyond histologic grade, and feature visualization revealed relevant histopathologic patterns. Conclusions: A foundation model-based MIL approach provides robust and interpretable prognostic modeling in HCC. DL-based prediction offers complementary information beyond conventional clinicopathological variables. Future efforts integrating multimodal data and improving generalizability will be essential for clinical translation. Full article

Automated cellular detection using deep learning is a key strategy for optimising cervical cancer screening by reducing the healthcare workload and inter-observer variability. However, analysing Whole Slide Image (WSI) patches presents challenges such as annotation scarcity, morphological complexity, and class imbalance. This study systematically evaluates YOLO11-n, YOLO11-s, and YOLO11-m to assess the impact of target variable granularity and training paradigms on performance. Four strategies were analysed: independent and multi-class models, each evaluated at both the specific cell label and diagnostic macro-group levels. To ensure clinical robustness, patient-level data partitioning was implemented to prevent data leakage. Performance was measured using precision, recall, and mAP (0.5 and 0.5:0.95). The results reveal critical trade-offs between fine-grained discrimination and model generalisation when varying the architectural complexity and labelling strategies. The findings indicate that diagnostic aggregation improves stability, whereas single-class training optimises specialised detection. These results provide methodological guidelines for designing AI-assisted screening systems and may inform future extensions of WSI-level diagnostic pipelines. Full article

Background/Objectives: While human epidermal growth factor receptor 2 (HER2) immunohistochemistry (IHC) is pivotal for breast cancer management, its reliance on additional tissue processing beyond routine H&E staining remains a clinical burden. Although virtual staining offers a potential solution, current methods often fail to explicitly account for HER2 score-specific expression patterns. To address this gap, we developed a score-aware framework designed for the precise generation of virtual HER2 IHC images. Methods: We introduce the non-contrastive multi-task (NCMT) framework, which integrates negative-free patch alignment, style–content constraints, and auxiliary HER2 score supervision for high-fidelity H&E-to-IHC translation. For rigorous evaluation, the model was validated on the BCI dataset, utilizing an official split of 3896 training and 977 independent test images derived from 51 whole-slide images. Results: NCMT demonstrated superior virtual staining performance, achieving a Fréchet Inception Distance (FID) of 38.8, a Kernel Inception Distance (KID) of 5.6, and an average Perceptual Hash Value (PHV) of 0.439. In downstream HER2 scoring tasks, while virtual IHC images alone yielded an accuracy of 83.01%, the fusion of H&E and virtual IHC further elevated performance to 97.85% accuracy and a 98.23% F1 score. These findings suggest that our framework effectively preserves diagnostic features while providing complementary information to H&E-based morphological analysis. Conclusions: NCMT enables HER2 score-aware virtual IHC generation from H&E and can serve as a complementary tool for HER2 assessment in digital pathology. Full article

Stain variability in digital pathology affects both cross-center diagnostic consistency and the robustness of downstream computational analysis. In this work, we formulate stain normalization as a variational inverse problem and derive a Screened Poisson Normalization (SPN) model from the steady-state reaction–diffusion mechanism underlying histological staining. In the CIE $L^{*}{a}^{*}{b}^{*}$ space, the model couples a gradient-domain fidelity term with a chromatic anchoring term, yielding a screened Poisson equation that preserves tissue morphology while enforcing color consistency. We prove that the corresponding variational problem is well-posed in $H^1\left(\mathsf{\Omega }\right)$ and stable with respect to perturbations of the input data. We further show that the screening term induces an intrinsic localization length ${𝓁}_c\sim {\lambda }_c^{-1/2}$, so that boundary perturbations decay exponentially away from tile interfaces. Based on this locality, we develop a non-overlapping tiled DCT-based spectral solver for gigapixel whole-slide images, enabling consistent tile-wise stain normalization and seamless whole-slide reassembly without heuristic boundary blending. Experiments on multi-scanner, multi-protocol, and archival-fading pathology datasets show that SPN achieves stable stain normalization with competitive chromatic alignment and strong preservation of diagnostically relevant microstructure, particularly in full-slide and tiled reconstruction settings. Supplementary experiments on synthetic pathology-like images further support the robustness of SPN under controlled color perturbations and indicate good generalization across diverse staining variations. Full article

Histopathological evaluation of tissue sections remains the gold standard for the diagnosis, classification, and grading of breast cancer (BC). The widespread adoption of whole-slide imaging (WSI) has enabled the digitization of histological slides and facilitated the development of artificial intelligence (AI) approaches for computational pathology. In recent years, machine learning and deep learning (DL) algorithms have been increasingly investigated for the analysis of hematoxylin and eosin (H&E)-stained images, with potential applications in tumor detection, histological classification, prognostic stratification, and prediction of treatment response. This narrative review summarizes recent developments in AI-driven models applied to BC histopathology and discusses their potential role in supporting diagnostic and prognostic assessment. Several studies have demonstrated the promising performance of DL algorithms in tasks such as the detection of lymph node metastases, assessment of residual tumor after neoadjuvant therapy, and prediction of clinical outcomes from histopathological images. Emerging research has also explored the possibility of inferring molecular and biomarker information from histology images, although these approaches currently identify statistical associations rather than direct molecular measurements. Despite the rapid expansion of this research field, significant barriers remain before routine clinical implementation can be achieved. Key challenges include dataset bias, variability in staining and image acquisition, limited external validation across institutions, and the need for transparent and reproducible model development. In addition, the translation of AI-based systems into clinical practice requires compliance with regulatory frameworks governing software used for medical purposes, such as those established by the U.S. Food and Drug Administration. Overall, AI represents a promising research direction in computational pathology and may contribute to decision-support tools capable of assisting pathologists in the analysis of digital slides. Continued efforts toward methodological rigor, large multicenter datasets, and prospective validation studies will be essential to determine the future role of AI in BC histopathology. Full article

Background: Colorectal cancer (CRC) remains the third most prevalent malignancy globally, with approximately 154,000 new cases and 54,000 projected deaths anticipated for 2025. The recent advancement of foundation models in computational pathology has been largely propelled by task-agnostic methodologies that overlook organ-specific crucial morphological patterns that represent distinct biological processes that fundamentally influence tumor behavior, therapeutic response, and outcomes. Methods: In this study, we develop a novel, interpretable AI model, PRISM (Prognostic Representation of Integrated Spatial Morphology), that incorporates a continuous variability spectrum within each distinct morphology to reflect the principle that malignant transformation occurs through incremental evolutionary processes. PRISM is trained on 15 million histological images extracted from surgical resection specimens of 2957 patients. Results: PRISM achieved superior prognostic performance for five-year OS (AUC = 0.70 ± 0.04; accuracy = 68.37% ± 4.75%; HR = 3.21, 95% CI = 2.18–4.72; p < 0.0001 using multi-variate cox-proportional hazards model), outperforming existing CRC-specific methods by 15% and AI foundation models by ~23% accuracy. It showed sex-agnostic robustness (AUC Δ = 0.02; accuracy Δ = 0.15%) and stable performance across clinicopathological subgroups, with minimal accuracy fluctuation (Δ = 1.44%) between 5FU/LV and CPT-11/5FU/LV regimens, replicating the Alliance cohort finding of no survival difference between treatments. Conclusions: These results establish PRISM as a promising, interpretable tool for AI-driven prognostication, with potential for future extension to other cancer types and stages Full article

Artificial intelligence (AI) in cardiovascular pathology involves the use of computational models, including machine learning and deep learning (DL), to analyse complex and heterogeneous data. These data include histopathological whole-slide images, cardiovascular imaging techniques such as cardiac magnetic resonance, echocardiography, computed tomography (CT), clinical parameters, and molecular information. The integration of these multimodal data sources allows AI to overcome the limitations of single-modality analysis, improving diagnostic accuracy, prognostic stratification, and personalised clinical decision-making while reducing inter-observer variability. Cardiovascular disease remains the leading cause of mortality worldwide, highlighting the need for more precise and timely diagnostic tools. AI has shown significant promise, particularly in digital pathology, where the digitisation of histological slides combined with advanced algorithms enables improved diagnosis, prognostic assessment, and translational research. This review summarises current AI applications in cardiovascular pathology, focusing on heart transplant rejection, cardiomyopathies, myocarditis, and atherosclerotic and valvular diseases. Automated methods offer important advantages, including diagnostic standardisation, quantitative histological analysis, and improved reproducibility. However, several challenges remain, such as the need for large, well-annotated shared datasets, limited interpretability of AI models, and ethical and legal issues related to clinical implementation. AI represents a promising tool for advancing cardiovascular pathology and personalised medicine, although robust multicentre validation is required before routine clinical adoption. Full article

Cervical cytology screening remains dependent on manual analysis, which is time-consuming and subject to variability. This study proposes a leakage-free hybrid deep learning framework for multiclass classification of cervical cells extracted from whole-slide Papanicolaou images. A fine-tuned DenseNet121 feature extractor was combined with three classifiers: Support Vector Machine (SVM), Stacked Extreme Learning Machine (SELM), and Cascaded Deep Forest (CDF). Experiments were conducted on the CRIC Cervix Collection dataset using slide-level data partitioning and group-aware stratified 7-fold cross-validation. Model comparison followed a paired non-parametric protocol (Friedman test with Wilcoxon post hoc and Holm correction). DenseNet121 + CDF achieved the highest cross-validation Accuracy (0.7370 ± 0.0357), significantly outperforming SVM (0.6644 ± 0.0287) and SELM (0.6431 ± 0.0471) (χ²(2) = 11.14, p = 0.0038; Kendall’s W = 0.79). Independent testing showed competitive generalization across models. These results support the statistical robustness of the Cascaded Deep Forest-based hybrid architecture for multiclass cervical cytology classification under realistic slide-level conditions. Full article

Background: Predicting genomic alterations from routine hematoxylin and eosin (H&E) whole-slide images (WSIs) may help triage molecular testing. Methods: We retrospectively enrolled 437 patients at Osaka Metropolitan University Hospital across 26 cancers, matched with clinical gene-panel data. We curated 1023 binary endpoints across SNV, CNV, and SV categories. We extracted slide embeddings from five pathology foundation models (Prism, GigaPath, Feather, Chief, and Titan) using a unified feature extraction pipeline and benchmarked them using a lightweight downstream Multi-Layer Perceptron (MLP) classifier. Using the best-performing patch feature system, we trained a multi-instance learning model to assess incremental benefit. Results: Titan achieved the highest and most stable transfer performance, with a median endpoint-wise Area Under the Receiver Operating Characteristic curve (AUROC) of 0.77 in the slide benchmarking; at the patch-level, prediction of APC_SNV reached an AUROC of 0.916, and prediction of KRAS_SNV reached an AUROC of 0.811 on the held-out test set. Conclusions: In a heterogeneous clinical gene-panel setting, pathology foundation models can provide strong baseline genomic-prediction signals without additional fine-tuning. We propose a practical, deployment-oriented two-stage workflow: rapid slide-embedding screening to prioritize robust representations and candidate endpoints, followed by patch-level training for high-value tasks where additional performance gains and interpretable regions are clinically worthwhile. Full article