Modern Pathology

Pathologic diagnoses of hematopoietic diseases require immunohistochemistry (IHC) stains selected by pathologists upon preview of H&E-stained slides. This multi-step workflow can delay diagnostic turnaround time by days. Hence, we developed the Hematopathology Automatic Triaging System (HATS), which automates IHC panel ordering directly from H&E whole-slide images using pretrained pathology foundation model representations combined with attention-based multiple-instance learning. After the most comprehensive evaluation of pathology foundation models for hematologic malignancy classification to date, encompassing seven publicly available models, we trained HATS on 4,996 whole-slide images from 1,607 patients spanning the ten most common lymphoma diagnostic categories. HATS achieves 84% case-level subtype classification accuracy (0.962 ROC-AUC), translating to 92% IHC panel ordering accuracy. In a blinded reader study, HATS outperforms practicing pathologists at predicting lymphoma subtypes from morphology alone (85% vs 65%). In an independent real-world validation of 230 clinical cases, after directing 7 cases with scant tissue for manual review, HATS-ordered IHC panels were sufficient for diagnosis in 72.6% of cases. By automating the triaging step while preserving full pathologist oversight, HATS offers a safe and practical entry point for clinical AI adoption in pathology.

Cervical cytology screening relies on the visual assessment of nuclear and cellular morphology across thousands of cells per slide, a labor-intensive task that could benefit from quantitative computational support. Here we present CytoLens, a modular deep learning system that detects and counts nuclei, delineates nuclei and whole cells, measures the nuclear-to-cell area ratio (N:C), classifies cells from multifocal images and scores candidate abnormality. We trained and tested each module on a public dataset matched to its task, without pooling datasets. CytoLens detected nuclei with an F1 score of 0.931 (95% confidence interval (CI), 0.914-0.946) and counted 75 of 100 fields to within one nucleus. Point-prompted segmentation achieved mean Dice coefficients of 0.901 for nuclei and 0.847 for whole cells and, without retraining, transferred to a second dataset in which measured N:C increased with lesion grade in parallel with expert outlines. Multifocal hierarchical classification reached a mean accuracy of 0.947 across four Bethesda-aligned categories, and abnormality scores ranked annotated abnormal candidates with an area under the receiver operating characteristic curve (AUROC) of 0.865. These results support the evaluation of CytoLens on patient-linked clinical specimens.

Background: Deep-learning systems for Gleason grading are developed almost entirely on high-end clinical scanners and on cohorts from a small number of Western institutions, yet deployment increasingly involves other devices and other populations. These two distribution shifts, device and population, are rarely tested together on the same physical slides. The PAR dataset, from Erbil, Iraq, digitizes each biopsy on three scanners and provides three distinct pathologist grades, so it permits both tests at once on a Middle Eastern cohort. A concurrent study by the dataset originators validated a task-specific model and two foundation models on PAR; we complement it by testing an inde-pendently developed detect-then-grade pipeline and by separating scanner effects on detection from scanner effects on grading. Methods: We applied one fixed model de-veloped on North American and European material to all 1017 whole-slide images (339 slides from 185 patients, three scanners; 49.6% clinically significant cancer) with no scanner-specific or population-specific tuning. We measured cancer detection (area under the ROC curve of the predicted cancer-tissue fraction), all-slide ISUP agreement of the deployed detect-then-grade pipeline (quadratic-weighted kappa, QWK), and grading agreement on pathologist-confirmed cancers, at the slide level and, because a case carries up to two slides, at the patient level. The reference reader was S.A.; thresholds and operating points were cross-validated leave-one-out; scanners were compared by paired within-biopsy bootstrap and confidence intervals confirmed by patient-cluster bootstrap. Results: Detection was statistically equivalent across scanners (AUC 0.987 to 0.991; paired differences at most 0.003) and transferred to this non-Western cohort with no per-population tuning. At a 95% sensitivity operating point the deployed pipeline reached cross-validated all-slide QWK of 0.86, 0.81, and 0.86 (Grundium, Hamamatsu, Leica), matching the inter-pathologist ceiling of 0.81, against 0.23 to 0.62 for the ungated model. Grading of confirmed cancers was scanner dependent: the compact Grundium (0.63) did not differ from the clinical Leica (0.67; paired difference 0.04, 95% CI -0.03 to 0.11), while both exceeded Hamamatsu (0.44). Results held at the patient level, with grading somewhat lower for two scanners; the two slides of a case disagreed in grade in 43% of cases, and patient clustering did not widen the intervals. Conclusions: Can-cer-tissue fraction is a triage signal robust across scanner and transferable to an un-derrepresented population for detection, while grading is the scanner-sensitive step. Prostate grading models should be deployed as a detect-then-grade pipeline, with grading validated per device and confirmed on the local population.

Background Histological diagnosis of early-stage mycosis fungoides (MF) is hindered by profound overlap with benign inflammatory dermatoses (BIDs), leading to diagnostic delays and extensive ancillary testing. We developed MIMIC (Multiple Instance-learning for Identification of Mycosis fungoides In Cutaneous biopsies), a weakly supervised deep learning model designed as a triage tool at initial H&E whole slide image (WSI) review to distinguish classic patch and plaque stage MF from BIDs. We externally validated the model and evaluated its clinical utility. Methods In this retrospective multicentre study, we trained a base model using weakly supervised attention based multiple instance learning on 3,339 WSIs from two Dutch centres. Crucially, all MF training labels were derived from a deeply phenotyped national cohort featuring strict multidisciplinary expert panel consensus diagnoses (the clinical gold standard). Transportability was evaluated on 371 WSIs from four independent European centres. A blinded reader study on 171 WSIs compared morphology only performance of MIMIC with 11 (dermato-)pathologists. We then retrained an updated model on all retrospective multicentre data and assessed clinical utility in a strictly held out, consecutive Utrecht cohort (2022-2023; 486 accessions, 863 WSIs). Primary analysis focused on classic MF versus BIDs (453 accessions). Decision curve analysis, using Platt scaled probabilities to correct for spectrum bias, evaluated net benefit at a prespecified, safety oriented threshold of 0.04. Findings The base model showed good multicentre transportability (mean centre specific AUROC 0.91; pooled AUROC 0.84). In the reader study, MIMIC achieved an AUROC of 0.87, exceeding the mean pathologist AUROC (0.79) and the best individual reader (0.83). In the consecutive MF versus BID cohort, the updated model achieved an AUROC of 0.87 (95% CI 0.81-0.92). At the 0.04 threshold, sensitivity was 97.8% (44/45 MF cases) and specificity 50.2%, reducing unnecessary ancillary workups by 39.9 per 100 screening cases versus a test all strategy. Interpretation By identifying nearly half of BIDs as low risk while preserving near complete sensitivity for classic early stage MF in a European digital pathology workflow, this unimodal H&E approach offers a scalable digital solution to reduce defensive ancillary testing and accelerate the diagnostic journey for patients with MF. Further validation is needed in non European centres and in populations with darker skin phototypes.

Cervical cancer is largely preventable when ab- normal cells are detected early. However, manual Pap-smear screening remains labor-intensive because a specimen may con- tain thousands of cells, borderline abnormalities are subtle, and expert cytology review is not equally accessible. Although artificial intelligence can reduce this workload, conventional classifiers often return a label without directly exposing the evidence used and may exploit staining or image-acquisition shortcuts. To address these limitations, we present CERVEX, a pathology-foundation-model framework with an additive class- evidence readout. Each class score is computed as the spatial mean of its class-specific evidence map plus a learned bias; therefore, the heatmap and numerical score are generated by the same forward calculation. CERVEX also produces a numerical research report containing class probabilities, evidence strength, ranked candidate regions, and segmentation-derived morphology, including N:C ratio and nuclear geometry. Under repeated slide- grouped evaluation on RIVA, with labelled target-cohort cells included during training, CERVEX distinguished abnormal from normal cells with AUROC 0.911 (SD 0.009), sensitivity 0.869, and specificity 0.796. Furthermore, fixed-grid analysis without cell coordinates distinguished low- from high-grade disease across 101 graded abnormal slides with AUROC 0.823 [0.728, 0.911]. On SIPaKMeD, nucleus and complete-cell segmentation reached Dice 0.932 and 0.942, while nuclear area fraction and N:C ratio reached measurement correlations of 0.977 and 0.962, respectively. Moreover, CERVEX produced class-specific spatial evidence while a separate held-out CRIC comparison did not resolve an accuracy difference from the conventional classifier, demonstrating computation-linked explainability with only 1.9% measured runtime overhead. Ultimately, CERVEX enables auto- mated analysis of cervical cytology microscope fields by linking cell classification, computation-linked spatial evidence, and in- terpretable numerical morphology indices in one pipeline. This supports transparent candidate-region review and establishes a practical foundation for future whole-slide screening and prospective clinical validation.

Background: General-purpose large language models (LLMs) are increasingly evaluated in diagnostic pathology, but prior studies have largely emphasized diagnostic accuracy rather than how models fail. We evaluated four LLMs for diagnostic performance, pathology-relevant errors and hallucinations, their burden, and potential clinical impact across multi-organ pathology cases. Design: In this retrospective multicenter study, 153 pathology cases from two institutions spanning 20 organs were evaluated using ChatGPT-5.3 (LLM1), Gemini 3 (LLM2), Grok 4.20 (LLM3), and Claude Opus 4.6 (LLM4). Each LLM received multi-magnification histologic images with clinical context and generated a microscopic description and diagnosis. No data splitting, training, or fine-tuning was performed. Twenty-one pathologists assessed 612 outputs for diagnostic correctness, error and hallucination type and burden, clinical impact (0-4), and overall performance (1-5). Results: Strict diagnostic accuracy was 48.9% overall (60.1% including partially correct diagnoses) and ranged from 36.6% to 58.2% across LLMs. Errors occurred in 82.0% of outputs and hallucinations in 76.6%; 90.2% contained at least one error or hallucination. Misinterpretation was the most frequent error (72.1%), while fabricated histologic features were the dominant hallucination type (74.5%). LLM2 had significantly lower misinterpretation rates than the other three LLMs, while LLM3 had significantly higher fabricated-feature hallucination rates than LLM1 and LLM2. Errors were more frequent in incorrect than correct diagnoses (97.5% vs 66.2%), as were hallucinations (95.1% vs 59.9%; both p < 0.001). Every strictly incorrect diagnosis contained an error and/or hallucination, while 79.9% of strictly correct diagnoses also contained at least one. In multivariable analysis, LLM3 was independently associated with higher odds of an incorrect diagnosis. Diagnostic correctness was the strongest determinant of high clinical impact; each one-point increase in hallucination burden increased the odds by 63% (adjusted OR, 1.63). Conclusion: Diagnostic accuracy alone substantially underestimates the safety limitations of general-purpose LLMs in histopathology. Errors and hallucinations were common even when the final diagnosis was correct, and their burden was independently associated with clinical impact. Evaluation frameworks should therefore assess both diagnostic correctness and the reliability and potential consequences of accompanying generated content.

Current state of the art computational pathology foundation models attain satisfying accuracy most general pathology tasks including detection of pancreatic ductal adenocarcinoma (PDAC) but fail to provide sufficient explainability, operating as "black boxes". Some interpretability effort has been made, with models providing attention maps that project back on the slide the models high attended areas that pushed for the given prediction result, without further naming or reasoning explanation which can limit pathologists trust toward the prediction. We developed C3PRO, a cell-centric framework attempting to provide deeper understanding and explainability, by using embeddings extracted using a pathology foundation model, from cell centered tiles containing both the cell and its close neighborhood, which were then clustered into recurrent morphological phenotypes. A spatial graph was additionally constructed on top of the back-projected vocabulary to account for recurrent motifs. The evaluation of the model performance involved a cohort of 809 patients (825 diagnostic biopsies) using nested cross validation. C3PRO was benchmarked against four different multiple instance learning frameworks and achieved on-par performance for diagnostic (0.958 AUC), survival (0.617) and resectability (0.696) prediction, while providing additional insights compared to the strongest baseline models. Each clustered phenotype was probed and extracted as a representative tile mosaic and annotated by two expert pathologists, constituting a phenotype vocabulary available for statistical analysis. These phenotypes were scored by pathologists with physical brightfield microscopy and replicated the independent statistical significance impact on overall survival prediction from diagnostic biopsies.

Background: Chronic inflammation is the most common histological feature in Benign Prostatic Hyperplasia (BPH), and T cells are a key component of immune infiltrate. Advanced BPH is commonly associated with the formation of nodules, but it remains unclear whether a link exists among T cell infiltration, nodular development, and BPH progression. Using a Toxoplasma gondii (T. gondii) model and human specimens, we characterize the subtypes of T cells present during prostatic hyperplasia and their association with nodular development of the prostate. Methods: Male CBA/j mice were intraperitoneally infected with T. gondii parasites, and flow cytometry was performed on the prostate to quantify the number of CD4+ and CD8+ T cells. Histology was used to score microglandular hyperplasia (MGH), and immunofluorescence was used to quantify and examine the locality of CD4+ and CD8+ T cells and compared that to human BPH tissue. Results: We found that infecting male mice with T. gondii resulted in an increase of both CD4+ and CD8+ T cells in the prostate acutely and that CD8+ cells remained sustained at chronically. We also established the presence of glandular nodule formation at this timepoint through hematoxylin and eosin (H&E) staining. Immunofluorescence revealed that CD8+ cells were found proximal to forming glandular nodules relative to non-nodular glands. We also found more CD8+ cells localized to non-nodular glands in nodular BPH tissue versus non-nodular BPH tissue. Finally, we discovered a higher prevalence of CD8+ cells in T. gondii IgG+ patients than in IgG- patients. All T. gondii IgG+ patients exhibited nodular BPH, whereas all but one IgG- patient exhibited non-nodular BPH. Conclusions: This study is the first to investigate the presence and location of CD4+ and CD8+ T cells within nodular and non-nodular BPH glands. We found an association of the presence of CD8+ T cells with nodular progression. This association held true in human prostate tissue. Translationally, CD8+ T cells may enhance nodular BPH progression, and T. gondii infection may promote this CD8+ T cell-mediated response.

Accurate, reproducible interpretation of kidney allograft biopsies is critical for diagnosis of graft injury to guide prognosis and management. The international Banff classification is a consensus diagnostic system based on semiquantitative histological lesion scoring on either extent or severity of kidney transplant biopsies. However, pathologist scoring is limited by substantial interobserver variability, constrained scalability, and the inherent nature of the scoring system itself. Here we present BanffNET, a weakly supervised, probabilistic deep learning framework that combines self-supervised feature extraction with a novel Bayesian multiple-instance learning framework to predict (continuously) the full spectrum of Banff lesion scores directly from whole-slide images (WSIs). Using lesion-specific aggregation functions tailored to localized (modeling lesion severity) and diffuse pathologies (modeling lesion extent), BanffNET generates interpretable, patch-level probability maps and calibrated slide-level scores. BanffNET's performance was assessed relative to consensus, biological correlates of rejection and clinical outcome, demonstrating superior consistency, transportability and generalization. Trained on 7,249 WSIs from three cohorts, BanffNET demonstrates consistent performance on 11,028 WSIs across five external test sets, performing on par or exceeding expert consensus across lesions. BanffNET scores align more closely than pathologist Banff scores with molecular profiles of rejection, offering a transparent, biologically grounded framework for computational pathology with relevance beyond transplantation.

Background & AimsHaematoxylin and eosin (H&E) staining remains the diagnostic gold standard for solid cancers, including hepatocellular carcinoma, and is increasingly complemented by genomic profiling for precision medicine. Inferring genomic alterations directly from H&E images could streamline testing, but heterogeneity and biases in human training data limit interpretation of genotype-phenotype associations. Here, we aimed to relate histologic to genomic pathology to provide biological explainability for mutation prediction models and assess the impact of germline variation on model performance. MethodsWe analysed 597 murine liver tumours with matched whole-genome sequencing and histopathology (163,835 image tiles; 22.9 million nuclei). Our controlled in vivo design accounted for germline variation, biological sex, and causal mutagen (N-diethylnitrosamine), removing confounding factors present in human cohorts. We trained and evaluated deep learning and supervised machine learning models to predict germline variation and cancer driver alterations from H&E. ResultsModelling accurately predicted germline and somatic alterations from histology, at both locus-specific and genome-wide scales. Quantitative image analysis revealed an unexpected association between Egfr driver mutations and hepatic steatosis, linking genotype to an interpretable morphological phenotype. While model performance declined when applied to tumours from unrepresented genetic backgrounds, this limitation was biologically informative, revealing strain-dependent differences in tumour evolution, notably the prevalence of whole-genome duplication. ConclusionsMachine learning integration of histological and genomic pathology enables accurate, interpretable inference of genetic alterations from H&E, potentially reducing reliance on costly ancillary molecular assays. Our predictions are supported by human-interpretable biological features, addressing concerns around "black-box" technologies. However, caution is required when applying such methods to samples with a genetic background that, even if closely related, is beyond the genetic horizon of training data.

Banff acute lesion scores underpin histologic classification of kidney allograft biopsies; however, biomarker studies rely on second-order associations with diagnostic categories that introduce confounding. We quantified the first-order relationships between Banff acute lesion scores and the validated urinary cell three-gene rejection signature. In 354 biopsy-urine pairs, three-gene signature scores computed using a locked regression equation incorporating absolute copy numbers of CD3E mRNA, CXCL10 mRNA, and 18S rRNA in urinary cell RNA-were related to glomerulitis (g), peritubular capillaritis (ptc), interstitial inflammation (i), and tubulitis (t). Signature scores rose monotonically with Banff acute lesion severity, with 1.5 to 1.8-fold higher odds of more severed g, ptc, i, and t (all P<0.0001), and showed good calibration. Associations remained robust for composite microvascular (g+ptc) and tubulointerstitial (i+t) indices and were strongest for severe g and t, supporting this signature as a noninvasive, quantitative readout of acute rejection pathology with immediate diagnostic applicability.

Background and Objective: Micro-ultrasound (MicroUS) is an emerging high-resolution imaging modality for prostate cancer diagnosis and real-time biopsy guidance, but its automated analysis is limited by labor-intensive expert annotation. We aimed to develop and evaluate a foundation model for data-efficient MicroUS image analysis. Methods: We developed muFM, a MicroUS-specific foundation model pretrained using DINOv2-style self-supervised learning on 1,736,391 unlabeled images from 780 patients. The model was fine-tuned and evaluated in three downstream cohorts for prostate capsule segmentation (public, 75 patients), urethra segmentation (institutional, 105 patients), and prostate cancer lesion localization (institutional, 190 patients). Under limited annotation, muFM was compared with randomly initialized models and foundation models pretrained on natural, medical, and ultrasound images. Evaluation used the Dice scores and 95th-percentile Hausdorff distance (HD95) for segmentation and free-response receiver operating characteristic analysis for lesion localization. Key Findings and Limitations: muFM improved performance across all three tasks, with the largest gains under limited annotation. With 10% of training scans, muFM achieved Dice scores of 93.8% for prostate capsule segmentation and 58.4% for urethra segmentation, versus 87.9% and 47.0% for the strongest comparators. Corresponding HD95 values were 2.0 and 4.3 mm. For lesion localization, muFM outperformed other models with approximately 0.75 sensitivity at two false-positive detections per 3D volume. Limitations include retrospective single-institution pretraining and lack of explicit 3D modeling. Conclusions and Clinical Implications: MicroUS-specific self-supervised pretraining improved data-efficient image analysis across prostate and urethra segmentation, and lesion localization. This approach may reduce annotation requirements for developing automated tools in MicroUS image analysis.

Foundation models (FMs) have rapidly become dominant in artificial intelligence and are increasingly being adopted in computational pathology. Numerous pathology-specific FMs have been developed and evaluated for a variety of downstream tasks, most commonly using frozen image embeddings with linear probes. While a pathology FM, at least implicitly suggests broad reusability across tasks, the transferability of these models to neurodegenerative disease-related tasks remains largely unexplored. In this study, we evaluated fourteen frozen feature extractors, including general-vision and pathology FMs, across four pathology image classification datasets spanning two specialized neurodegenerative disease immuno-histochemistry (IHC) tasks (Tau neurofibrillary tangle (NFT) and amyloid-{beta} plaque classification) and two cancer-related hematoxylin and eosin (H&E) tasks (the breast tissue BACH dataset and the multi-tissue TIL dataset). The cancer-related H&E datasets represent tasks more closely aligned with the predominant pretraining domain of many current pathology FMs, whereas the neurodegenerative IHC datasets represent a more specialized domain with limited representation in existing FM pretraining cohorts. Frozen embedding linear probes were compared against a conventional supervised ResNet-50 convolutional neural network (CNN) trained directly on image tiles using identical train, validation, and test splits. Across both neurodegenerative IHC datasets, the supervised CNN substantially outperformed all frozen FM linear probes. In contrast, pathology FMs achieved performance comparable to, and in some cases exceeding, the supervised CNN across the two cancer-related H&E datasets, with UNI2-h achieving the highest performance on BACH and several pathology FMs performing on par with the CNN on the larger TIL dataset. Furthermore, a supervised CNN trained using only 1% of the Tau NFT training data (1,938 tiles) still exceeded the performance of the best frozen FM linear probe trained on the complete dataset. Together, these results suggest the transferability of frozen pathology FMs may depend strongly on how well the downstream task is represented by their pretraining domain. These findings demonstrate frozen pathology FMs transfer effectively to the cancer-related H&E tasks evaluated here but may be less effective for specialized neurodegenerative IHC tasks less well represented in current FM pretraining cohorts. Expanding pathology FM pretraining datasets to include a broader range of disease domains and staining modalities may therefore improve transferability to specialized pathology applications. Our findings also highlight the continued importance of conventional supervised learning and motivate future work investigating nonlinear probes and end-to-end FM fine-tuning.

Muscle-invasive bladder cancer (MIBC) carries high risks of relapse and death despite contemporary care. In this study we aimed to evaluate whether the pretreatment immune microenvironment varied with HER2 status and if the differences influenced the response to neoadjuvant chemotherapy (NAC) and survival. We conducted a retrospective study of 62 treatment-naiive MIBC patients diagnosed on pretreatment transurethral resection of bladder tumor (TURBT) specimens collected between 2008 and 2024. All patients received NAC followed by radical cystectomy. HER2 immunohistochemistry (IHC) was scored 0/1+/2+/3+ using gastroesophageal criteria. PD-L1 was assessed by combined positive score (CPS). CD3, CD20, and CD68 cell densities were quantified by digital image analysis. Overall survival was measured from the date of TURBT. HER2 IHC scores were 2+/3+ in 28 of 62 tumors (45%) and 0/1+ in 34 (55%). All seven IHC 2+ tumors lacked ERBB2 amplification by fluorescence in situ hybridization (FISH). NAC response occurred in 11 of 34 tumors with HER2 IHC 0/1+ (32%) and 9 of 28 tumors with HER2 IHC 2+/3+ (32%) (P = 1.00). PD-L1 expression was not associated with HER2 status. Immune-cell analyses showed trends toward higher CD3+ T-cell density, CD20+ B-cell density, and TM ratio in NAC responders, but no significant differences by HER2 status or combined PD-L1/HER2 subgroup status. After Bonferroni correction for multiple comparisons, none of the immune-cell metrics remained statistically significant. Overall survival did not differ between the HER2 IHC groups (P = 0.71). These findings indicate that HER2 IHC 2+/3+ expression is present in a substantial subset of MIBC but is not associated with NAC response or overall survival in this cohort. Further study is warranted to define the clinical relevance of MIBC with HER2 IHC 2+/3+ and its potential role in HER2-directed therapeutic strategies.

Background: Clinical registries and institutional quality control (QC) in pathology heavily depend on complex histopathological data. However, critical tumor characteristics are mostly locked in unstructured free-text reports. We hypothesize that locally deployed, resource-efficient language models operating within a secure institutional network can reliably extract relevant variables from German reports of hepatocellular carcinoma (HCC), thereby enabling automated data curation without transferring sensitive patient data. Methods: We developed a local, privacy-preserving open-weight Large Language Model (LLM) pipeline consisting of two consecutive stages: Stage 1 utilizes a lightweight gatekeeper model to isolate the target cohort of confirmed Hepatocellular Carcinoma cases from a screening pool of N = 2,864 reports. Stage 2 deploys a high-capacity model to extract 15 predefined histopathological variables, including tumor-node-metastasis staging, histological grading, vascular invasion, and selected immunohistochemical features into structured JSON format. For both Stages, pipeline performance was evaluated using Accuracy, Macro F1-scores, and Cohen's kappa compared to expert-curated reference annotations. Results: The top-performing Stage 1 gatekeeper isolated the target patient cohort with zero classification errors, effectively reaching complete agreement with the reference standard (ground truth) and correcting inaccurate administrative metadata. For Stage 2 feature extraction, current-generation models demonstrated robust structural reliability, reducing the burden of manual post-processing. The 31B model configuration (Gemma 4) virtually eliminated objective factual errors, achieving near-perfect agreement with the reference standard (Macro F1 > 0.99). Conclusion: Local two-stage pipelines offer a highly accurate, resource-efficient, and secure solution for clinical text processing. This framework serves as a privacy-preserving engine for institutional quality control and the scalable curation of pathology registries.

Convolutional neural networks (CNNs) have shown strong capabilities for image analysis. However, deploying these models in medical settings is complicated by their limited transparency. Over recent years, many approaches have been developed to overcome the so-called "black box" problem of deep neural networks. Here, we show how such explainable AI (XAI) approaches can be applied to not only improve transparency but also training data quality and model performance, with classification of lung damage in histopathology images as use case. First, we conducted a thorough exploratory data analysis and visually compared the compressed multi-dimensional representations of the histology images from the last CNN layers with labels given by pathologists to reveal flaws in the training data. Second, we used Gradient-based Class Activation Mapping (Grad-CAM) as well as SHapley Additive exPlanations (SHAP) values to identify image regions that significantly contributed to the model decisions. To overcome identified shortcomings, we then finetuned additional top layers of the CNNs and introduced model architectures with attention which improved model performance. In summary, we developed a practical workflow that uses interpretability analyses to examine model perception, assess label consistency, and guide model refinement on the example of lung histopathology scoring, demonstrating how XAI approaches can increase both transparency and model performance. The code for this paper is shared at https://github.com/Aitslab/Histology_XAI.git.

BackgroundTraditional immunohistochemistry (IHC) with chromogen detection has limited multiplex capacity, detecting at most 4 protein markers per tissue section simultaneously, thereby restricting comprehensive spatial analysis of valuable human biopsies. We developed and validated a robust serial IHC (sIHC) staining method to detect multiple antigens on a single kidney biopsy slide, maximizing data yield for diagnosing and studying complex kidney diseases. MethodsFormalin-fixed, paraffin-embedded kidney biopsy sections were subjected to repeated IHC/imaging cycles with antibody removal using an optimized sodium dodecyl sulfate-glycerol buffer stripping protocol. Images were then co-registered, and analysis was performed using a variety of methodologies, including color deconvolution, cell segmentation, and spatial clustering. ResultsThis optimized sIHC method successfully detected up to 20 antigens on a single slide. Combining image analysis and artificial intelligence software, for example with HALO (Indica Labs), the assay assembles high-dimensional images and enables quantitative histology and single-cell spatial analysis. Using this advanced method, we were able to identify rare cell populations, such as double-negative T cells, that are challenging to detect conventionally. ConclusionWe have developed a validated, high-capacity sIHC protocol that uses standard IHC procedures with commercially available, clinically validated off-the-shelf antibodies. This method is a valuable, cost-effective tool for obtaining extensive, high-dimensional single-cell-resolved spatial data from limited pathology samples, such as a human kidney biopsy.

Background Pre-implantation biopsy utility for deceased-donor kidney acceptance is controversial due to processing limitations and interpretive variability. We evaluated whether systematic histological assessment, including expert evaluation and quantification with the BanffNET automated deep learning system (17 lesions), predicts post-transplant failure under optimal laboratory conditions. Methods Biopsies from deceased donor kidneys (N=733) underwent optimized laboratory processing, whole slide imaging, expert Banff 2024 pathologic assessment, and BanffNET lesion quantification. We assessed their incremental predictive value beyond clinical characteristics for early graft dysfunction, longitudinal eGFR/UPCR trajectories, need for indication biopsies, and death-censored graft failure. Results Histopathology improved clinical model discrimination by 1-4% and increased explained variance for longitudinal trajectories by 1-2%. Continuous bivariate surface analysis, however, revealed that while histology adds marginal overall predictive accuracy, mapping continuous composite and individual BanffNET scores against the Kidney Donor Risk Index (KDRI) uncovers distinct topological risk gradients for early graft dysfunction and long-term graft failure. Conclusion While pre-implantation biopsies add minimal overall prognostic value, continuous bivariate mapping of KDRI and BanffNET scores uncovers distinct synergistic risk gradients for graft failure. These continuous clinical-histological surfaces could optimize organ salvage by precisely identifying viable high-KDRI kidneys lacking critical compounded damage and vice versa.

1Accurate subtype diagnosis is essential for guiding therapy and predicting patient outcome in malignant mesothelioma. Most computational pathology models are trained on large tissue images from resection specimens, which maximises information for training but limits model relevance in real-world diagnostic settings where small biopsies are the most usual tissue source. In this work, we assembled a large multicentre cohort of HES- and HPS-stained mesothelioma biopsy slides. We used a self-supervised learning model to evaluate the associations of biopsy-driven morphology patterns with histological subtype, molecular markers, and survival. The discovered histomorphology patterns captured a continuum of tissue phenotypes spanning epithelioid, sarcomatoid, and non-tumour morphologies. Also, patient-level HPC representations achieved excellent performance for distinguishing epithelioid from non-epithelioid mesothelioma (AUC = 0.94) and demonstrated predictive value for immunohistochemistry (IHC) markers. Additionally, HPC-derived features alone achieved performance comparable to established clinical and molecular variables (C-index = 0.65), while integration of HPCs with clinical and marker information improved performance to a C-index of 0.69. Several HPCs were significantly associated with favourable or adverse prognosis and reflected known subtype-specific biological patterns. In conclusion, self-supervised learning can discover interpretable histomorphological phenotypes directly from routine mesothelioma biopsies without further training. These AI-derived phenotypes capture clinically and biologically relevant information, linking tissue architecture to molecular characteristics, histological subtypes, and patient outcomes. The proposed framework provides a thorough evaluation of real-world biopsy data using a pre-trained model, without the need for computationally intensive retraining, and addresses the question of whether SSL-based AI can be deployed out of the box in clinical settings.

SignificanceRare-cell identification in fluorescence microscopy remains challenging because targets are sparse and background varies between specimens. Combining specimen-specific fluorescence enrichment with image classification may enable efficient and more specific automated detection of rare cells. AimWe developed a two-stage framework to identify and quantify candidate rare circulating hybrid neoplastic cells (CHCs, ECAD+/CD45+) in peripheral blood mononuclear cell (PBMC) preparations from tumor-bearing and tumor-naive mice. ApproachPBMCs from 28 mice were imaged by multichannel fluorescence microscopy. Matched unstained samples established animal-specific ECAD and CD45 background distributions for candidate cell enrichment. Blinded multi-annotator consensus labels were used to train a convolutional neural network (CNN) from DAPI, ECAD, and CD45 image crops. Generalization was evaluated by leave-one-animal-out validation across 10 random initializations. Final classification used a 10-model ensemble, and rare-cell burden was compared between groups using negative-binomial regression with total segmented-cell count as an exposure. ResultsOf the 1,065,512 segmented cells, enrichment retained 10,176 candidates (0.96%), reducing the search space by >99%. Four of five evaluable tumor-bearing animals showed reproducible held-out discrimination, with median quantified area under the receiver operator characteristic curve (AUROCs) of 0.918-0.951; one animal was a reproducible outlier (median AUROC, 0.338). Ensemble deployment identified 157.94 positive-consensus cells per 50,000 segmented cells in tumor-bearing animals versus 49.55 in controls. The estimated rare-cell rate was 3.15-fold higher in tumor-bearing animals (95% CI, 0.91-10.99; two-sided p=0.071; prespecified one-sided p=0.036). ConclusionsSpecimen-specific fluorescence enrichment combined with supervised image classification reduced the cellular search space and enabled automated quantification of a rare CHC (ECAD+/CD45+) phenotypes. Cross-animal validation also identified specimen-specific generalization failure, highlighting the importance of biological-specimen-level validation.