Large B-cell lymphoma (LBCL) with IRF4 rearrangement (LBCL- IRF4 -R) is a rare subtype predominantly diagnosed in children and young adults. Whether adult LBCL cases with IRF4 rearrangement ( IRF4 -R) should be classified as LBCL- IRF4 -R remains unclear. Clinicopathological and molecular features of 61 adult LBCL cases with IRF4 -R were analyzed and compared to diffuse large B-cell lymphoma, not otherwise specified (DLBCL, NOS), to assess biological heterogeneity. The 61 cases grouped by patient age and site were classified as follows: Group 1 included 13 patients aged ≤40 years, whose features supported LBCL- IRF4 -R, showing favorable outcomes with high frequencies of IGH::IRF4 fusion and IRF4 mutations. Group 2 comprised 37 patients aged >40 years with tumors at usual sites; 17 early-stage cases largely retained LBCL- IRF4 -R characteristics, whereas three other early-stage cases showed molecular or clinical features more consistent with DLBCL, NOS with IRF4 -R. Twelve advanced-stage cases showed aggressive behavior with adverse DLBCL, NOS-associated mutations (e.g., TP53 ), suggesting classification as follicular lymphoma grade 3A (FL-3A) and DLBCL with IRF4 -R or DLBCL, NOS with IRF4 -R. Five cases lacked sufficient data for definitive classification. Group 3 included 11 patients aged >40 years with tumors at unusual extranodal sites; except for one case classified as LBCL- IRF4 -R, the remaining 10 cases showed aggressive clinical behavior with frequent MYD88 mutations, favoring classification as FL-3A and DLBCL with IRF4 -R or DLBCL, NOS with IRF4 -R. These findings support a multidimensional approach that integrates age, tumor site, and clinicomolecular features to refine classification, enhance risk stratification, and guide personalized management in adult LBCL cases with IRF4 -R.

This study aims to identify the role of tumor immune microenvironment (IME) in ductal carcinoma in situ (DCIS) in predicting benefit of whole breast irradiation (WBI). Different subtypes of tumor infiltrating lymphocytes (TILs), tumor associated macrophages (TAMs) and tertiary lymphoid structures (TLSs) were determined in tumor tissues of DCIS cohort who received breast-conserving surgery (BCS). In total, 165 patients were enrolled with 113 received WBI. After a median follow-up of 73.7 months, 15 ipsilateral breast tumor recurrence (IBTR) events occurred. Nine IBTRs occurred outside of the original quadrant (elsewhere failure event, EFE). After LASSO and multivariate Cox regression analyses, the ER negative status, high ratios of CD4 + /CD8 + , dense CD68 + TAMs, sparse CD8 + T cell and dense TLSs with follicular dendritic cells (F-TLSs) remained independent risk factors of IBTR (all p < 0.05) to develop a nomogram. Points of nomogram were defined as immune microenvironment score (IMS) which divided all patients into low-, intermediate- and high-risk groups. Significant differences in IBTR existed among these three risk subgroups (5y-rate: 0.0

Vision foundation models (VFMs) have emerged as a powerful technological breakthrough to address this fundamental challenge. Representing a new paradigm for universal and efficient image representation, VFMs are pre-trained on vast quantities of unlabeled data to learn a foundational understanding of visual features. Due to their strong generalization capabilities and remarkable data efficiency, they have become a key development direction in CPath, laying a solid foundation for a new generation of AI applications. This paper provides a systematic review and comprehensive analysis of representative VFM construction methods specifically designed or adapted for the CPath domain. We trace the technical evolution and core architectural features, starting from earlier CNN-based models (like ResNet) pre-trained on ImageNet, and culminating in the current state-of-the-art based on the vision transformer (ViT) architecture. A primary focus is placed on the pivotal role of self-supervised learning (SSL), which empowers these models to learn from the data itself, bypassing the need for manual annotations. We analyze the distinct strategies employed, including contrastive learning (e.g., MoCo, SimCLR) which maximizes feature similarity between augmented views of the same patch, and masked image modeling (MIM). MIM methods, such as MAE and the approach used by DINO v2, are particularly well-suited to the structured and often redundant textures of histopathology, as they train models to reconstruct masked or corrupted patch tokens, forcing them to learn rich local context. This review emphasizes the excellent transferability and significant performance gains these SSL-driven VFMs achieve across diverse downstream pathology tasks, such as cell classification, tissue segmentation, and patient survival prediction, often outperforming fully-supervised models trained on limited data. Furthermore, this paper critically analyzes the key challenges that current pathology VFMs must overcome for practical clinical deployment. A major difficulty lies in the structured representation of the unique multi-scale resolution of WSIs. These gigapixel images contain diagnostic information spanning from the cellular level (high-magnification) to the global tissue architecture (low-magnification). VFMs, often based on ViTs, typically process fixed-size patches in isolation. This creates a "contextual gap", as simple aggregation of patch-level features fails to capture the long-range dependencies and hierarchical relationships crucial for diagnosing complex pathologies. Another significant hurdle is the complex morphological heterogeneity and domain shift introduced by real-world variability, such as different scanners, preparation protocols, and H&E staining variations, all of which demand exceptional model robustness. Finally, we discuss the flexibility of task adaptation. The sheer size of VFMs (often containing billions of parameters) makes full fine-tuning computationally prohibitive and prone to overfitting on small downstream datasets. We explore the critical role of parameter-efficient fine-tuning (PEFT) techniques, such as adapters and prompt-tuning, as a necessary solution for efficiently steering these models toward specific clinical tasks without catastrophic forgetting. Finally, this review looks forward to promising future development directions and research opportunities. The most critical need is the development of multi-modal VFMs that build a holistic patient model by fusing WSI data (morphological phenotypes) with genomics (genotypic drivers) and structured clinical records. Additionally, for any clinical tool, interpretability and explainability are non-negotiable. Future models must not only be accurate but also provide trustable, human-readable outputs, such as localizing predictive regions via attention maps, to gain acceptance from pathologists. Lastly, the optimization of model efficiency, through methods like quantization and knowledge distillation, will be paramount for deploying these massive models in resource-constrained hospital IT infrastructures.

Folate receptor-α is an ideal precision therapy target of ovarian cancer. The standardization of FRα assay and interpretative criteria is essential for ensuring diagnostic consistency and enhancing clinical efficacy in therapeutic applications. This study aims to analytically verify and evaluate the clinical performance of the VENTANA FOLR1 Assay. This real-world study of Chinese patients analyzed FRα expression using the VENTANA FOLR1 RxDx assay in 313 samples from diverse anatomical sites. Inter- and intra-observer agreement in FRα scoring was evaluated, and correlations between FRα positivity and clinicopathological parameters were examined. Three pathologists demonstrated excellent inter- and intra-observer agreement (> 97 %) in FOLR1 interpretation. 40.9 % of cases showed high FRα expression, with a significantly higher positivity rate in high-grade serous carcinoma among the Chinese cohort. Primary tumors exhibited higher FRα positivity than metastatic lesions (44.2 % vs 32.2 %, p = 0.04). Chemotherapy exposure did not significantly alter FRα positivity across ovarian, fallopian tube, and primary peritoneal cancers, remained comparable to that of the overall cohort (41.2 % vs 40.9 %). Excision/resection samples were identified as optimal for FRα assessment. Our findings demonstrate the high reliability of the VENTANA FOLR1 Assay in Chinese clinical settings. Additionally, we conducted a systematic investigation into the associations between FRα expression and clinicopathological characteristics, highlighting its capacity to reflect FRα heterogeneity, maintain stability in post-chemotherapy FRα expression across various tumor types, and achieve robust performance in excision/resection samples. These findings underscore the value of standardizing FRα testing to improve patient selection for FRα-targeted MIRV therapies in China.

Obesity exacerbates rheumatoid arthritis (RA). However, the underlying mechanisms remain incompletely defined. Elucidating these mechanisms can help the identification of novel therapeutic targets. Herein, we used high-fat diet (HFD)-induced obese collagen-induced arthritis (CIA) mice to investigate these mechanisms. Immunohistochemistry revealed that obesity exacerbated joint inflammation and cartilage degradation. Next, integrated label-free quantitative proteomics and cytometry by time-of-flight (CyTOF) were used to characterize lymphocyte subsets. Proteomic profiling identified 26 differentially expressed proteins in obese versus lean CIA mice, including the transcription factors EOMES and KLF2, the TGFβ receptor (TGFβR) signaling component TGFBR2, and the tissue-resident memory (TRM) T cell marker CD103. CyTOF analysis revealed a robust 3.0-fold increase (P = 0.0043) in the proportion of CD103⁺ TRM cells among CD3⁺ T cells in obese CIA mice, characterized by a large effect size. Immunofluorescence results confirmed this increase in synovial tissues. Treatment with asiaticoside (a TGF-β/Smad-suppressing triterpenoid) significantly reduced TRM cell proportions (P < 0.05) and ameliorated symptoms in obese CIA mice. Collectively, these findings establish a novel mechanistic axis in which obesity-induced TGFβR-hyperactivation promotes TRM cell accumulation, which exacerbates arthritis severity in this RA model. Our findings provide a preclinical rationale for targeting TGFβR/TRM in human RA with obesity as a comorbidity.

PURPOSE:POU2F3 is a newly identified immunohistochemical marker specific for the chemosensory tuft cell-related subtype of small cell lung cancer (SCLC) (SCLC-P). The characteristics of SCLC-P remain incompletely defined, and POU2F3 expression patterns across different organs and tissue types are poorly documented. MATERIALS AND METHODS:We assessed POU2F3 expression in 253 SCLCs, with comprehensive clinicopathological and genomic characterization of POU2F3-positive tumors. POU2F3 expression profiles were investigated in other major lung cancer types (n = 2537) and other tumors across different organs and tissue types (n = 195). RESULTS:POU2F3 was expressed in 10.28% (26/253) of all SCLC cases and was strongly associated with low expression of standard neuroendocrine (NE) markers (Syn, CgA, CD56, and INSM1). In NE-low/negative SCLC and NE-high SCLC, the POU2F3 positive rates were 83.33% (20/24) and 2.62% (6/229), respectively. Additionally, POU2F3 was detected in squamous cell carcinoma (2.35%) and large cell NE carcinoma (25%) but was negative in lung adenocarcinoma, NUT carcinoma, large cell carcinoma, pleomorphic carcinoma, SMARCA4-deficient thoracic undifferentiated tumor, atypical carcinoid, and adenoid cystic carcinoma. Notably, the highly heterogeneity of POU2F3 expression was observed in 7.3% (3/41) of surgical SCLC specimens. In extrapulmonary tumors, POU2F3 was positive in 37.5% (6/16) of extrapulmonary small cell NE carcinomas, 16.67% (5/30) of thymic tumors, 1 of 2 extrapulmonary large cell NE carcinomas, and 1 of 1 nasopharyngeal carcinoma. SCLC-P tends to be more prevalent in surgical specimens (P = .009) and earlier TNM stage (P = .045). Next-generation sequencing revealed that SCLC-P (n = 6) exhibited enrichment in MYC gene amplification and lower mutation rate of RB1 but similar rates of TP53 and PTEN alterations as POU2F3-negative SCLC (n = 13). CONCLUSIONS:This study establishes POU2F3 as a critical diagnostic biomarker for NE-low/negative SCLC, demonstrating high specificity in distinguishing SCLC-P from other thoracic malignancies and small blue round cell tumors. We delineate the distinct clinicopathological and genomic profile of POU2F3-driven SCLC (SCLC-P), providing a foundation for its diagnostic application. Further validation in expanded cohorts is warranted to confirm its clinical utility.

Foundation and multimodal models now match expert-level performance on many diagnostic, prognostic, and biomarker tasks in pathology. Yet only a handful of AI systems are used in routine clinical practice, and the products that have reached patients differ substantially from research prototypes. We define this mismatch as the adoption paradox of computational pathology.We first survey the technical landscape from task-specific deep learning to large unimodal foundation models, multimodal systems, and early agentic architectures. We then examine what has actually entered the clinic, identifying four product archetypes (digital pathology platforms, population scale cytology screening, assistive detection in surgical pathology, and quantitative immunohistochemistry scoring). Using a three stage maturity model algorithmic capability (Stage 1), system integration (Stage 2), and institutional adoption (Stage 3), we analyze the structural barriers that gate each transition.Three interconnected barriers explain most of the gap: (1) Data and infrastructure fragility [(pre-analytical variability, scanner-induced domain shift, format fragmentation, annotation scarcity, manual quality control (QC)]; (2) Workflow misalignment (cognitive rhythm of pathologists, automation bias, scenario-dependent latency); (3) Institutional trust deficits (shallow interpretability, incomplete prospective validation, unclear reimbursement, unsettled liability, and regulatory gaps for generative/adaptive systems).We outline system-level pathways for each stage, including infrastructure first, AI, workflow-embedded intelligence, and adaptive governance. Our central claim is that the next phase of progress will depend less on architectural novelty than on the slower institutional work that turns capability into clinical benefit. The framework provides an actionable lens for regulators, developers, and healthcare organizations to diagnose why a given AI system remains a prototype and what is needed to move it into routine use.

PurposeCurrent daily usage of Trastuzumab Deruxtecan (T-DXd) is guided by immunohistochemistry (IHC)-based HER2 assessment, with known inconsistency and inaccuracy to differentiating IHC 0 from 1 +. In this study, a quantitative HER2 assay based on the Quantitative Dot Blot (QDB) method was explored to fill this unmet need.MethodsConsecutive resection specimens of HER2 IHC 0 and 1+ from invasive breast cancer patients were assigned to training (n=106) and validation cohorts (n=119), respectively by admission time. Protein lysates were extracted from 2x5 μm FFPE slices for HER2 quantification while the adjacent slice was used for IHC staining.ResultsQDB was demonstrated to be more consistent than IHC with an inter-rater Intraclass Correlation Coefficient (ICC) of 0.877 (95%CI: 0.840-0.908) vs. 0.513 (95%CI: 0.433-0.601). Receiver Operating Characteristic (ROC) analysis was performed benchmarked with unanimous agreement of 18 pathologists in the training cohort to achieve an Area Under the Curve (AUC) of 0.9477 (p<0.005). A cutoff of 0.2746 nmole/g was also identified with its imprecision interval to stratify specimens into 0 ( C95), with overall concordance of 90.9% and 87.3% when benchmarked with unanimous and consensus agreement (≥75%) of pathologists in the training cohort, and 92.0% and 90.5% when validated double-blinded with 12 pathologists in the validation cohort. More importantly, even among specimens unanimously categorized as IHC 0, there were ~15% specimens classified as 1+ by QDB.ConclusionOur results supports the QDB HER2 assay as an alternative option to guide T-DXd daily usage by distinguish Her2-low from Her2 0 while setting the stage for outcome-based patient stratification.

Background: Clear cell renal cell carcinoma (CCRCC) is the predominant subtype of renal cell carcinoma and is characterized by frequent chromosome 3 alterations, including 3p deletion, monosomy, and aneuploidy. However, the absence of standardized fluorescence in situ hybridization (FISH) cut-off values has led to inconsistent reported frequencies and limited clinical integration of this accessible assay. This study aimed to establish clinically applicable cut-off values, propose a practical three-tier classification, and evaluate its diagnostic accuracy and clinicomolecular correlation with tumor aggressiveness. Methods: FISH using VHL (3p25.3) and CEP3 probes was performed on 1748 RCC cases (1655 CCRCC, 48 papillary RCC, 45 chromophobe RCC). Cut-off values were determined by combining ROC analysis with Youden's index and the mean + 3SD method from normal renal tubular cells. A paired cohort of 97 CCRCC cases with targeted next-generation sequencing was stratified into three subgroups (3p intact, isolated 3p loss, broad chr3 change) for clinicomolecular comparison, including 3D principal component analysis. Results: Clinically applicable thresholds of 30% for 3p deletion and 20% for monosomy identified chromosome 3 alterations in 76.9% of CCRCC cases. The combination of both markers achieved superior diagnostic accuracy (AUC = 0.82). Aneuploidy was significantly associated with higher WHO/ISUP grade (p < 0.001) and older age (p = 0.006). The three-tier classification showed stepwise progression of aggressive features (older age, higher grade, larger tumor size) and increasing PBRM1 mutation frequency from the 3p intact to the broad chr3 change group. Conclusions: This study establishes standardized FISH cut-offs and a practical three-tier classification that captures a continuous spectrum of genomic instability and tumor aggressiveness in CCRCC. Routine 3p FISH provides a simple, cost-effective, and reproducible tool with substantial diagnostic and stratification value that complements more complex genomic profiling.

This study investigated the associations between tumor-infiltrating lymphocytes (TILs), genomic features, and prognosis in HER2+ early breast cancer (EBC) patients receiving adjuvant trastuzumab. We retrospectively analyzed 864 HER2+ EBC patients from Shanghai Ruijin Hospital (2009-2017). The optimal threshold of TILs for predicting disease-free survival (DFS) and overall survival (OS) was explored. Whole-exome sequencing (WES) on 261 tumors assessed the mutational profiles, tumor mutational burden (TMB), and copy number alteration (CNA). Associations between these genomic features, TIL levels, and prognosis were further evaluated. TILs showed a right-skewed distribution (median: 15%, IQR: 1-30%), and higher TIL levels were significantly associated with hormone receptor negativity and high histologic grade (P < 0.001). A 15% TIL threshold optimally predicted prognosis, with low-TIL (≤15%, 63.0%) patients showing inferior DFS (HR: 1.63, P = 0.009) and OS (HR: 2.12, P = 0.037). WES identified frequent mutations in TP53 (62.8%), PIK3CA (34.5%), and BRCA2 (9.6%). A higher TIL density was observed in TP53-wild-type, low-TMB or low-CNA tumors (P < 0.05). PIK3CA mutations conferred a significant DFS advantage. Integrating TIL level with PIK3CA or BRCA2 mutational status yielded distinct DFS trajectories (log-rank P = 0.023 and 0.040, respectively); patients with both high TIL levels and either PIK3CA or BRCA2 mutations had the most favorable outcomes. Stromal TILs at a 15% cutoff provide robust prognostic information in trastuzumab-treated HER2+ EBC. Integrating TIL levels with PIK3CA or BRCA2 mutational status enables refined risk stratification, offering a practical framework for personalized treatment decisions.

Eosinophilic vacuolated tumor (EVT) is an emerging renal entity and further studies are required to characterize this neoplasm. EVT is considered generally indolent because metastasis or death from the disease has never been reported. Herein, we report the first case of EVT with mediastinal lymph node metastasis to validate its at least borderline behavior. Based on the morphological features, we additionally used a large panel of immunohistochemical antibodies and next-generation sequencing (NGS) to confirm the diagnosis of EVT and exclude other renal entities. This case report highlights the notion that EVT can exhibit mediastinal lymph node metastasis and patients with EVT need to be closely followed.

ELOC -mutated renal cell carcinoma ( ELOC -RCC), a newly recognized tumor entity in the fifth edition of the WHO Classification of Tumors of Urinary and Male Genital Organ Tumors (5th WHO Classification), presents morphologic and immunohistochemical (IHC) features overlapping those of clear cell RCC (ccRCC), RCC with fibromyomatous stroma (RCC-FMS), and clear cell papillary renal cell tumor (ccPRCT). Confirmation of an ELOC mutation is required for a definitive diagnosis. This study aims to enhance the understanding of ELOC -RCC's morphologic and molecular characteristics and to develop an affordable and practical panel for its preliminary differentiation based on morphologic and IHC features. Representing one of the largest cohorts of ELOC -RCC, this research involved a retrospective analysis of 56 suspected cases at Shanghai Ruijin Hospital from January 2022 to March 2024, identifying 15 cases through next-generation sequencing (NGS). We report an ELOC mutation site (c.274G>A, p.Glu92Lys), which has not been previously reported in the literature. NGS analysis also showed recurrent mutations in MAP2K4 and HRAS in ELOC -RCC, though their implications are not yet clear. In addition, we describe a case of ELOC -RCC with a PARP4 mutation. Our findings indicate that the "basally polarized" nuclear arrangement and the "apical/apicolateral polarized" staining patterns of CD10 and EMA offer valuable diagnostic clues for differentiating ELOC -RCC from low-grade ccRCC. Furthermore, the immunophenotypic profile of CD10+/AMACR+/GPNMB- appears helpful for differentiating ELOC -RCC from both ccPRCT and mTOR pathway-mutated RCC-FMS ( mTOR -RCC-FMS). However, genetic testing remains indispensable, as evidenced by one CK7-negative ELOC -RCC case.

Esophageal basaloid squamous cell carcinoma (EBSCC) is a rare subtype of esophageal squamous cell carcinoma (ESCC) that might be misdiagnosed or missed in clinical practice. Through RNA sequencing on 20 pure EBSCC and 11 poorly differentiated ESCC samples, we identified CSPG4 as a potential marker for EBSCC at the mRNA level, and verified with CSPG4 immunohistochemical staining in these cases. Then we assessed the value of CSPG4 expression in differential diagnosis and prognosis in EBSCC cases (n = 360) and conventional ESCC cases (n = 160) from 11 different institutions with whole-tissue sections. CSPG4 had acceptable discriminatory capacity for EBSCC, with an AUC ROC of 0.891. The optimal cutoff value for the H-score determined by ROC curve analysis was 77.5, with 82.7% sensitivity and 79.9% specificity. In well-moderately differentiated ESCC with CSPG4 expression, staining was mostly observed on the edge of the tumor nest or infiltration front, with different expression pattern from EBSCC. The H-scores of CSPG4 expression for the EBSCC component were higher than those for other components in the same tissue section (P < 0.05). The expression level of CSPG4 was similar in groups with different percentages of the EBSCC component (P > 0.05). In additional 505 ESCC patients, patients with high CSPG4 expression had decreased DFS and OS, especially in stage I-II disease (P < 0.001). The similar prognostic significance was also found in EBSCC. Our data indicate that CSPG4 is not only a sensitive but also a specific marker for the diagnosis of EBSCC. CSPG4 might also be a prognostic marker for ESCC.