Stain transfer in computational pathology offers a potential alternative to traditional immunohistochemical (IHC) staining. Existing methods primarily focus on image-to-image feature learning, relying heavily on generative adversarial networks and contrastive learning. However, textual information associated with pathology images, such as staining protocols or expression profiles, has not yet been incorporated into stain transfer frameworks. This omission limits the model’s ability to capture intrinsic characteristics and to generalize across different staining protocols. In this paper, we propose a text-driven stain transfer model that leverages both textual descriptions and stain channel information to guide the learning process. Our approach enhances feature comprehensiveness and discriminative power while preserving fine-grained structural details. Experiments on both public and private datasets demonstrate that the proposed model outperforms state-of-the-art (SOTA) GAN-based virtual staining approaches across multiple evaluation metrics.

Aims Traditional breast cancer treatment relies on binary human epidermal growth factor receptor 2 (HER2) classification, while HER2-low breast cancer represents an intermediate subgroup. Given the efficacy of novel antibody-drug conjugates, this study aimed to clarify whether HER2-low breast cancer is biologically and clinically distinct from HER2-0 disease, especially in response to neoadjuvant chemotherapy. Methods A retrospective analysis was performed on 1274 early breast cancer patients receiving neoadjuvant chemotherapy. Patients were divided into HER2-0 (n=210), HER2-low (n=590) and HER2-positive (n=474) groups. Clinicopathological features, pathological complete response (pCR) and long-term survival were compared. Results HER2-low tumours were predominant in Luminal A-like subtype, with lower pCR rate (14.7%) than HER2-0 (27.6%) and HER2-positive (35.5%) groups. Hormone receptor (HR)-positive status was associated with higher pCR in HER2-low patients. HER2-low tumours showed higher lymph node metastasis and lower Ki-67 levels. No significant differences in disease-free survival (DFS) or overall survival (OS) were observed between HER2-low and HER2-0 groups, with distinct prognostic factors for each subgroup. Conclusions HER2-low breast cancer exhibits unique clinicopathological features and differential chemotherapy response. However, it shows similar survival to HER2-0 disease with standard neoadjuvant chemotherapy. The clinical value of HER2-low status may rely on targeted antibody-drug conjugate therapy.

Background:Endocrine therapy (ET) is the primary treatment for hormone receptor (HR)-positive breast cancer. Based on the 2023 expert consensus, the Chinese Consensus Group on Breast Cancer Endocrine Therapy updated this consensus by integrating clinical research data and practical experiences. Methods:(I) Establishment of expert group: the expert group consists of experts from departments such as medical oncology, breast surgery, and pathology. (II) Literature search: mainly conducted in English databases (such as PubMed, Embase, and Cochrane Library) with a search cutoff date of Sep 30th 2025. (III) Assessment of evidence quality and recommendation strength: evidence quality and recommendation opinions are graded based on the evidence category and recommendation level of the Chinese Society of Clinical Oncology (CSCO) guidelines. Results:The 2026 consensus updated clinical research data on ET for both early and advanced breast cancer, defined the eligible population for neoadjuvant ET, and clarified the application of adjuvant cyclin-dependent kinase 4/6 inhibitor (CDK4/6i) therapy in moderate and high-risk patients. For advanced breast cancer, the consensus optimized the treatment strategy of first-line ET plus CDK4/6i, and recommended preferred post-CDK4/6i regimens for patients with distinct clinic-pathological characteristics. It also provides clear recommendations on the indication, optimal timing, and methodology of genetic testing. Conclusions:This consensus may provide clinical guidance for the standardized application of ET, thereby facilitating precise and individualized treatment for HR-positive breast cancer.

Pathological assessment guides lung cancer diagnosis, treatment selection, and prognostic evaluation, yet current CPath approaches rely on task-specific models for isolated objectives. Although pan-cancer foundation models offer versatility, they lack subspecialty-level depth and have not been evaluated across clinical workflows or prospectively validated in real-world settings. We introduce PulmoFoundation, a multi-center, prospectively validated, randomized controlled trial (RCT)-evaluated foundation model for comprehensive lung pathology assessment across pre-operative, intra-operative, and post-operative care. Built upon Virchow2 via subspecialty-specific pretraining using 40,000 diagnostic H E-stained whole-slide images (WSIs), PulmoFoundation was systematically evaluated on 26,000 WSIs across 32 clinically relevant tasks. In addition to accurately predicting molecular markers and patient survival, our model achieves clinical-grade performance in core diagnostic tasks across biopsy, frozen section, and surgical resection slides. In a registered prospective study of 1,357 patients across 11 diagnostic tasks, our model achieved an average AUC of 92.3

Objective: Accurate detection of PIK3CA mutations is essential for guiding PI3K-targeted therapies in breast cancer, yet sequencing is not universally accessible, and single-modality prediction models have limited performance. This study developed a multimodal deep learning framework integrating whole-slide imaging (WSI) and structured clinical data to improve mutation prediction. Methods: A total of 1,047 patients from TCGA and 166 patients from 3 external centers were included. The histopathology model used a transformer-based pretrained encoder (H-optimus-0) and a clustering-constrained attention multiple instance learning (CLAM-SB MIL) classifier to generate WSI-level representations. The clinical model incorporated engineered clinical variables and an extreme gradient boosting (XGBoost) model. A decision-level late fusion strategy (Multimodal PIK3CA Model, MPM) combined probabilistic outputs from both branches. Performance was evaluated with the area under the curve (AUC) and secondary metrics. Interpretability was assessed via attention heatmaps and shapley additive explanations (SHAP) analysis. Results: MPM outperformed single-modality models. It achieved an AUC of 0.745 on TCGA and maintained stable performance across external cohorts (0.695, 0.690, and 0.680). SHAP analysis identified molecular subtype as the most influential clinical feature, whereas attention maps highlighted mutation-associated morphological regions. Conclusions: The developed multimodal framework effectively integrates complementary morphological and clinical information, and provides a robust and generalizable method for predicting PIK3CA mutation status. Strong multicenter adaptability and biological interpretability support its potential use as a clinical decision-support tool and an accessible alternative to molecular testing.

Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal malignancies worldwide, and accurate prognostic prediction remains highly challenging due to its marked biological heterogeneity and complex tumor microenvironment. To address this challenge, a histopathomics-based survival prediction system (HPSurv) was developed using histopathological whole-slide images (WSIs) for individualized overall survival (OS) prediction. Within this framework, pathological tissue classification, quantitative characterization of tumor spatial heterogeneity, and a survival Transformer were integrated to enable multi-level representation learning from histopathological data. The system was developed and evaluated in 1020 patients across five independent cohorts. Compared with conventional clinicopathological indicators, significantly improved prognostic performance was achieved across multicenter cohorts (p < 0.05), with a mean C-index of 0.761 and time-dependent AUCs of 0.936, 0.877, and 0.772 for predicting 6-month, 2-year, and 3-year survival, respectively. Subgroup analyses further supported its role as an independent prognostic factor and suggested its potential utility in stratifying patients with respect to ACT-related outcomes. In addition, significant associations with key PDAC molecular pathways were observed, providing biological insights into the model predictions and supporting interpretability. In the study, an interpretable and high-performing artificial intelligence (AI) framework for quantitative modeling of PDAC was established. Objective characterization of tumor heterogeneity and accurate postoperative survival prediction are enabled, with potential value for personalized management in PDAC.

ABSTRACT The immune microenvironment of invasive mucinous adenocarcinoma of the lung (IMA), a rare and heterogeneous subtype, remains poorly characterized, limiting insights into its potential response to immunotherapy. In this multicenter study, we systematically evaluated programmed cell death ligand 1 (PD‐L1) expression (using tumor proportion score, TPS, and combined positive score, CPS) and cluster of differentiation 8‐positive (CD8+) tumor‐infiltrating lymphocyte (TIL) infiltration in the largest cohort to date of pathologically confirmed pure IMAs (n = 312), supported by single‑cell transcriptomic analysis. PD‐L1 positivity was low (TPS≥1%: 9.0%; CPS≥1: 28.5%). While PD‐L1 alone showed no prognostic significance, high CD8+ TIL percentage and density were independent, favorable prognostic factors for relapse‐free survival, particularly in patients not receiving adjuvant therapy. By integrating TPS and CD8+ TIL percentage, we established a novel four‐category immune phenotype classification that identified a distinct subgroup (Type IV: PD‐L1+/CD8+) with significantly better outcomes. Preliminary analysis of 20 patients who received immune checkpoint inhibitors suggested that Type IV patients may derive greater clinical benefit. Single‐cell RNA sequencing analyses revealed a paucity of effector CD8+ T cells in IMA. This work defines the unique immune landscape of IMA, introduces a clinically relevant immune phenotyping framework with prognostic and predictive potential, and provides a rationale for future immunotherapeutic strategies in this rare malignancy.

2597 Background: Definitive concurrent chemoradiotherapy (dCCRT) is considered the standard treatment for esophageal squamous cell carcinoma (ESCC). The PACIFIC study demonstrated that consolidation durvalumab significantly improves overall survival (OS) in patients with stage III non-small cell lung cancer (NSCLC) after dCCRT. However, the efficacy of consolidation immunotherapy in ESCC still remains unclear. We conducted a clinical trial to evaluate the efficacy of camrelizumab in patients with unresectable, locally advanced ESCC following dCCRT. Methods: This single-arm, phase 2 study enrolled patients with locally advanced ESCC. All participants received dCCRT with involved-field irradiation (IFI). Patients were treated with camrelizumab within 1 to 42 days after completing dCCRT. Camrelizumab was administered intravenously over 30 minutes every 2 weeks for up to 12 months. The primary endpoint was progression-free survival (PFS). Secondary endpoints included disease control rate (DCR), objective response rate (ORR), duration of response (DoR), overall survival (OS), and safety. Results: Thirty-five patients were enrolled between April 2020 and November 2023. Data from 32 patients were analyzed. As of December 22, 2024, the median follow-up was 25.1 months (IQR 5.5–56.8). Twelve patients experienced disease progression, and seven patients died. The DCR was 59.4%. The median PFS and OS were not reached. The 1- and 2-year PFS rates were 81.3% and 60.6%, respectively. The 1- and 2-year OS rates were 96.9% and 81.0%, respectively. The most common adverse events were grade 1-2. No grade 4 or 5 adverse events were reported. Pneumonia occurred in 31.3% of patients, all of whom experienced grade 1-2. Conclusions: Consolidative camrelizumab following definitive concurrent chemoradiotherapy with IFI shows promising efficacy and manageable toxicity in patients with unresectable locally advanced ESCC. Clinical trial information: NCT04286958 .

Pathology foundation models (PFMs) generalize well across computational pathology tasks but remain costly for gigapixel whole-slide image analysis. Here, we present LitePath, a deployment-friendly framework that addresses model over-parameterization and patch-level redundancy. LitePath combines LiteFM, a compact model distilled from Virchow2, H-Optimus-1 and UNI2 using 190 million patches, with the Adaptive Patch Selector for task-specific patch selection. Compared with Virchow2, LitePath uses 28x fewer parameters and 403.5x fewer FLOPs. On an NVIDIA Jetson Orin Nano Super, it processes 208 slides per hour, 104.5x faster than Virchow2, and consumes 0.36 kWh per 3,000 slides, 171x less energy than Virchow2 on an RTX 3090 GPU. We evaluated LitePath on 45 multicenter cohorts across four organs and 33 tasks, comprising 37 classification and 8 survival cohorts, including 33 internal, 10 external and 2 prospective cohorts, with 17,837 slides from 9,977 patients disjoint from the pretraining data. Among 22 PFMs, LitePath achieved the best average rank (6.56 vs. 6.58 for Virchow2), retained 99.71

Objective To develop an multimodal interpretable model that integrates whole slide images(WSI)and clinical features,and to validate its efficacy in predicting pathological complete response(pCR)in lung cancer patients following neoadjuvant therapy.Methods The clinicopathologic data who received neo-adjuvant therapy of patients as well as hematoxylin and eosin stained sections were retrospectively collected be-tween March 2015 and March 2025.For the WSI branch,the predictive performance of five pathology founda-tion models(CTransPath,Virchow2,H-optimus-0,Phikon-V2,and UNI-V2)was compared within the CLAM-SB framework to identify the optimal feature extractor.The clinical branch was constructed using the ex-treme gradient boosting(XGBoost)model.Subsequently,a multimodal prediction model for pathologic com-plete response(MP-pCR)was established through decision-layer logistic regression fusion.Model performance was evaluated using the area under the receiver operating characteristic curve(AUC),accuracy,sensitivity,specificity,F1-score,and Brier score.Interpretability analysis was performed using attention heatmaps and the Shapley Additive Explanations(SHAP)algorithm.Results A total of 728 patients were enrolled in this study.Among them,a total of 676 patients from the Fourth Hospital of Hebei Medical University were selected as internal dataset,which was randomly divided into training set(n=536),validation set(n=70),and in-ternal test set(n=70)at a ratio of 8∶1∶1.Additionally,32 patients from the Affiliated Hospital of Hebei University and 20 from Handan First Hospital were selected as external validation set 1 and external validation set 2,respectively.In the internal testing set,UNI-V2 emerged as the top-performing feature extractor for the WSI branch,achieving an AUC of 0.774(95%CI:0.688-0.861).The resulting MP-pCR multimodal model yielded an AUC of 0.812(95%CI:0.725-0.899)in the internal testing set,outperforming both the WSI-only model(0.774)and the clinical-only model(0.780),with an accuracy of 81.8%,sensitivity of 70.6%,specificity of 86.5%,and a Brier score of 0.125.In external validation set 1,the MP-pCR model achieved an AUC of 0.746(95%CI:0.598-0.898)and an accuracy of 75.0%;in external validation set 2,it achieved an AUC of 0.722(95%CI:0.538-0.918)and an accuracy of 80.0%.Attention heatmaps re-vealed that the model-focused regions were primarily concentrated within the tumor parenchyma and treatment-related peritumoral areas.SHAP analysis indicated that preoperative treatment regimen,pathological diagnosis,tumor-to-stroma ratio,age,and smoking history were the top five contributors to pCR prediction.Conclusions MP-pCR model developed and validated based on multicenter data outperforms single-modality models in pre-dicting pCR for lung cancer.The interpretability results demonstrate good consistency with clinicopathologic knowledge,suggesting that the model holds promise as a potential auxiliary reference for the clinical evaluation of treatment response and the optimization of individualized therapeutic decisions.

In locally advanced gastric cancer, a substantial number of patients relapse with liver metastasis months after an apparently curative operation, and standard tumor staging offers little warning of who is at risk. Here, we develop the Radiopathomics-Clinical Stratification Assessment (RCSA), an interpretable model that integrates three complementary sources of information: radiomic features from preoperative computed tomography, pathomic features from routine hematoxylin and eosin tumor slides, and conventional clinical features. Trained on patients from one hospital and then tested on separate internal, external, public, and prospective trial groups (NCT02555358), RCSA consistently separates high- and low-risk patients, with area under the curves between 0.862 and 0.909. Tumors it labels low-risk carry a notably more active immune environment, indicating that these patients are the ones most likely to gain from added immunotherapy. RCSA therefore turns existing hospital data into individualized guidance for postoperative follow-up and treatment. Accurate prediction of metachronous liver metastasis (MLM) after curative surgery for locally advanced gastric cancer (LAGC) remains challenging. Here, the authors develop a multimodal model to predict MLM in LAGC patients by integrating clinical, imaging, and pathology data across multi-centre cohorts, also revealing patients who could be more responsive to adjuvant immunotherapy.

Abstract Precise evaluation of cervical lymph node metastasis (CLNM) and genetic mutations (BRAFV600E/TERT promoter, TERTp) is pivotal for tailoring surgical and prognostic evaluation and adjuvant strategies in thyroid cancer (TC). Although current methods have limitations, we aim to develop deep learning (DL) models to predict CLNM and genetic mutations from TC frozen sections. We developed a DL framework using 2499 frozen‐section whole‐slide images from 2176 TC patients across five centers. The model was trained with a transfer learning‐based feature extractor and an attention‐based multiple instance learning (MIL) classifier, and validated on both internal and external cohorts. StyleGAN3‐based data augmentation was employed to tackle class imbalance for TERTp prediction, while interpretability was assessed via attention heatmaps and Leiden clustering. The CLNM prediction model achieved a patient‐level AUROC of 0.918 internally and 0.803–0.885 across three external validation datasets. For BRAFV600E prediction, AUROCs attained 0.814 internally and spanned 0.750––0.811 in external validation. In TERTp mutation prediction, GAN‐based augmentation increased the AUROC to 0.804 (internal) and 0.732 (external), up from 0.782 and 0.724, respectively. Attention maps visualized CLNM correlations with invasive tumor margins, while mutations localized to specific cellular morphology features. Our DL models accurately predict CLNM and genetic mutations from TC frozen sections, potentially reducing unnecessary procedures and providing a rapid alternative to traditional molecular testing.