Modern Pathology

Digital pathology enables large multi-centre studies of histological specimens, but differences in staining protocols and slide quality can compromise the comparability of quantitative results. We analysed 686 PSR-stained liver biopsies from four independent cohorts spanning more than 20 clinical sites to assess how stain variability affects automated fibrosis quantification and model uncertainty. A U-Net ensemble was trained to segment collagen and to estimate pixel- and tile-level predictive uncertainty. Across markedly heterogeneous staining conditions, the ensemble achieved strong segmentation performance (Dice 0.83-0.90) and produced informative uncertainty maps that identified artefacts and out-of-distribution regions. Epistemic uncertainty values were typically below 0.002, providing a practical criterion for flagging unreliable predictions. Our results demonstrate that ensemble-based uncertainty estimation complements stain-standardisation efforts by quantifying prediction confidence directly from model outputs, improving the reliability and interpretability of collagen proportionate-area measurements across multi-centre datasets. This framework supports more trustworthy and reproducible digital-pathology workflows for fibrosis assessment and other histological applications. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=111 SRC="FIGDIR/small/25326981v2_ufig1.gif" ALT="Figure 1"> View larger version (31K): org.highwire.dtl.DTLVardef@1781d1dorg.highwire.dtl.DTLVardef@bf83adorg.highwire.dtl.DTLVardef@15dff59org.highwire.dtl.DTLVardef@274874_HPS_FORMAT_FIGEXP M_FIG C_FIG HighlightsO_LIA retrospective cohort of liver biopsies collected from over 20 healthcare centres has been assembled. C_LIO_LIThe cohort is characterized on the basis of collagen staining used for liver fibrosis assessment. C_LIO_LIA computational pipeline for the quantification of collagen from liver histology slides has been developed and applied to the described cohorts. C_LIO_LIUncertainty estimation is evaluated as a method to build trust in deep-learning based collagen predictions. C_LI

A pathologist typically diagnoses tissue samples by examining glass slides under a light microscope. The entire tissue specimen can be stored digitally as a Whole Slide Image (WSI) for further analysis. However, managing and diagnosing large numbers of images manually is time-consuming and requires specialized expertise. Consequently, computer-aided diagnosis of these pathology images is an active research area, with deep learning showing promise in disease classification and cancer cell segmentation. Robust deep learning models need many annotated images, but public datasets are limited, often constrained to specific organs, cancer types, or binary classifications, which limits generalizability. To address this, we introduce the UCF multi-organ histopathologic (UCF-MultiOrgan-Path) dataset, containing 977 WSIs from cadaver tissues across 15 organ classes, including lung, kidney, liver, and pancreas. This dataset includes [~]2.38 million patches of 512x512 pixels. For technical validation, we provide patch-based and slide-based approaches for patch- and slide-level classification. Our dataset, containing millions of patches, can serve as a benchmark for training and validating deep learning models in multi-organ classification.

By generating realistic histologically-stained images from label-free image data, virtual staining (VS) methods have the potential to streamline clinical workflows, improve image consistency and provide new ways of visualizing and analysing tissues. This article describes a new VS approach based on the application of conditional generative adversarial networks to translate high-resolution phase images of unstained tissues, recovered using Fourier ptychographic microscopy (FPM), into brightfield H&E images. Compared to other label-free imaging methods, FPM offers unique advantages for VS as it allows simultaneous capture of sample amplitude and phase information, simplifying the pixelwise registration required for supervised training. FPM combines high spatial resolution with a large field of view and a large depth of field making it well suited to large format imaging of histological tissues. The method is readily implemented by modifying a conventional brightfield microscope using simple, low-cost optoelectronic hardware. Using colonic polyps as a test case, we compare FPM and VS-FPM images to brightfield whole slide images (WSIs) captured using a pathology slide scanner. Our results show that FPM images captured at 4x magnification have a spatial resolution equivalent to WSIs captured at 20x magnification. Virtual H&E images of unstained tissues generated using VS-FPM closely match brightfield images of the same tissue sections captured after chemical staining, enabling pathological assessment and diagnosis. HighlightsO_LIFourier Ptychography (FPM) can capture large-format complex images of tissue sections. C_LIO_LI4x FPM images and 20x images of H&E-stained tissues have equivalent spatial resolution. C_LIO_LIConditional GANs can generate accurate brightfield H&E images from FPM phase images. C_LIO_LIVS-FPM enabled pathologists to assess and diagnose unstained colonic polyps. C_LIO_LIThe method can be applied to other sample types in histo- and cyto-pathology. C_LI Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=59 SRC="FIGDIR/small/25327933v1_ufig1.gif" ALT="Figure 1"> View larger version (25K): org.highwire.dtl.DTLVardef@103fcaaorg.highwire.dtl.DTLVardef@c135d6org.highwire.dtl.DTLVardef@b5ad68org.highwire.dtl.DTLVardef@18baf36_HPS_FORMAT_FIGEXP M_FIG C_FIG

Multiclass segmentation of microanatomy in kidney biopsies is an important and non-trivial task in computational renal pathology. In a multicenter study, we densely annotated basic anatomical objects (glomeruli, tubules, and vessels) in 261 regions of interest of 147 kidney biopsy WSIs sourced from the archives of hospitals in Amsterdam, Utrecht, and Leiden (Netherlands). And we trained multiple UNet- and Mask2Former-based models on WSI-level and patch-level splitting methods, and compared their performance across training strategies. Test performance was assessed on 24 annotated renal WSIs from Leuven (Belgium) with sensitivity analysis on the extent of fibrosis and inflammation. At the patch-level, UNet-ResNet18 achieved comparable performances to M2F-Swin-B with average Intersection over Union of all classes (A-IoU, 0.84 vs 0.94), as well as per-class IoU. However, at the WSI-level, M2F-Swin-B significantly surpassed UNet-ResNet18 with large margins on A-IoU (0.84 vs 0.48), with similar observed in per-class IoU. Notably, M2F-Swin-B outperformed UNet-ResNet18 in scenarios characterized by a higher degree of fibrosis and inflammation (A-IoU, 0.76 vs 0.66). Furthermore, at the WSI-level, M2F-Swin-B achieved IoU score of arteries to 0.58, whereas UNet-ResNet18 only achieved 0.33. In this study, we found that the attention mechanism in Mask2Former enables visibly crisper and more uniform segmentation, particularly when data is inadequate. Mask2Former-based models outperform UNet-based models in challenging areas from inflamed and fibrotic renal biopsies.

Research using whole slide images (WSIs) of scanned histopathology slides for the development of artificial intelligence (AI) algorithms has increased exponentially over recent years. Glass slides from large retrospective cohorts with patient follow-up data are digitised for the development and validation of AI tools. Such resources, therefore, become very important, with the need to ensure that their quality is of the standard necessary for downstream AI development. However, manual quality control of such large cohorts of WSIs by visual assessment is unfeasible, and whilst quality control AI algorithms exist, these focus on bespoke aspects of image quality, e.g. focus, or use traditional machine-learning methods such as hand-crafted features, which are unable to classify the range of potential image artefacts that should be considered. In this study, we have trained and validated a multi-task deep neural network to automate the process of quality control of a large retrospective cohort of prostate cases from which glass slides have been scanned several years after production, to determine both the usability of the images for research and the common image artefacts present. Using a two-layer approach, quality overlays of WSIs were generated from a quality assessment undertaken at patch-level at 5X magnification. From these quality overlays the slide-level quality scores were predicted and then compared to those generated by three specialist urological pathologists, with a Pearson correlation of 0.89 for overall usability (at a diagnostic level), and 0.87 and 0.82 for focus and H&E staining quality scores respectively. We subsequently applied our quality assessment pipeline to the TCGA prostate cancer cohort and to a colorectal cancer cohort, for comparison. Our model, designated as PathProfiler, indicates comparable predicted usability of images from the cohorts assessed (86-90%), and perhaps more significantly is able to predicts WSIs that could benefit from re-scanning or re-staining for quality improvement. We have shown in this study that AI can be used to automate the process of quality control of large retrospective cohorts to maximise research outputs and conclusions.

Cutaneous squamous cell carcinoma (cSCC) poses significant clinical challenges due to its rising incidence and potential for metastasis. Histopathologic risk stratification is further limited by substantial inter-observer variability. Unsupervised AI approaches based on content-based image retrieval offer scalable and interpretable decision support for diagnostic pathology. The objective of this study was to evaluate the use of image retrieval within histopathology atlases to stratify cSCC tumor differentiation from whole-slide images (WSIs), while comparing different patch selection and feature extraction strategies. This retrospective study included 552 archived WSIs comprising 385 well-differentiated, 102 moderately differentiated, and 66 poorly differentiated cases collected across Mayo Clinic sites in Arizona, Florida, and Minnesota. Image atlases were constructed using multiple patch aggregation strategies (Mosaic, Collage, and Montage) and deep learning encoders (KimiaNet, PathDino, and H-Optimus-0). A leave-one-WSI-out evaluation framework was used to assess differentiation classification performance using accuracy, specificity, sensitivity, and F1 score. Mosaic combined with KimiaNet achieved the highest Top-1 accuracy (74.9%) and specificity (92.6%), while Mosaic with H-Optimus-0 yielded the best Top-5 accuracy (79.0%) and macro-F1 score (62.6%). Collage combined with KimiaNet produced the highest Top-5 specificity (99.5%). The generalizability of the evaluated AI models varied across hospitals, reflecting differences in imaging protocols, staining practices, and patient populations. Overall, unsupervised image search and retrieval provides effective, annotation-free support for cSCC differentiation and has the potential to enhance dermatopathology workflows when appropriate combinations of patch selection and feature ex-traction methods are employed.

BackgroundProstate cancer (PCa) is among the most common cancers in men and its diagnosis requires the histopathological evaluation of biopsies by human experts. While several recent artificial intelligence-based (AI) approaches have reached human expert-level PCa grading, they often display significantly reduced performance on external datasets. This reduced performance can be caused by variations in sample preparation, for instance the staining protocol, section thickness, or scanner used. Another limiting factor of contemporary AI-based PCa grading is the prediction of ISUP grades, which leads to the perpetuation of human annotation errors. MethodsWe developed the prostate cancer aggressiveness index (PCAI), an AI-based PCa detection and grading framework that is trained on objective patient outcome, rather than subjective ISUP grades. We designed PCAI as a clinical application, containing algorithmic modules that offer robustness to data variation, medical interpretability, and a measure of prediction confidence. To train and evaluate PCAI, we generated a multicentric, retrospective, observational trial consisting of six cohorts with 25,591 patients, 83,864 images, and 5 years of median follow-up from 5 different centers and 3 countries. This includes a high-variance dataset of 8,157 patients and 28,236 images with variations in sample thickness, staining protocol, and scanner, allowing for the systematic evaluation and optimization of model robustness to data variation. The performance of PCAI was assessed on three external test cohorts from two countries, comprising 2,255 patients and 9,437 images. FindingsUsing our high-variance datasets, we show how differences in sample processing, particularly slide thickness and staining time, significantly reduce the performance of AI-based PCa grading by up to 6.2 percentage points in the concordance index (C-index). We show how a select set of algorithmic improvements, including domain adversarial training, conferred robustness to data variation, interpretability, and a measure of credibility to PCAI. These changes lead to significant prediction improvement across two biopsy cohorts and one TMA cohort, systematically exceeding expert ISUP grading in C-index and AUROC by up to 22 percentage points. InterpretationData variation poses serious risks for AI-based histopathological PCa grading, even when models are trained on large datasets. Algorithmic improvements for model robustness, interpretability, credibility, and training on high-variance data as well as outcome-based severity prediction gives rise to robust models with above ISUP-level PCa grading performance.

AO_SCPLOWBSTRACTC_SCPLOWWith the rise of digital pathology, integrating digital slides with deep learning-based decision support systems is becoming increasingly common in clinical practice. Tissue region segmentation which is distinguishing tissue from background/artefacts, is an important pre-requisite in many digital pathology pipelines both for the laboratories as their first step in digitalizing the glass slides of tissue samples and turning them to whole slide images (WSIs) using scanners, and also for DL researches such as region-of-interest cropping, tumor detection, cell segmentation. However, it is well known that WSI scanners can fail in detecting all tissue regions, due to the tissue type, or due to weak staining and this is because of their not robust enough tissue detection algorithms which makes segmentation of WSIs a challenging task. Hence, this study develops a fast, lightweight, accurate, CPU-ready DL approach, enabling fast and reliable tissue region segmentation model by training and testing it across seven different institutional H&E and IHC stained WSIs to result a strong in generalization with the 22 to 56 s/WSI inference time using CPU that markedly outperforms classical OTSU thresholding, particularly in preserving challenging or faint tissue regions by achieving notably higher and more consistent performance than OTSU, with median Jaccard and Dice scores of approximately 0.86 and 0.92, respectively, compared to OTSU whcih was between 0.56 and 0.72. Our approach provides a practical, open-source solution for resource-limited pathology settings. We publicly released dataset obtained from Bahcesehir Medical School, and code to foster benchmarking and further advances in efficient, deployable computational pathology. The model could be used in digital slide scanners to improve the scanning process and in the pre-processing stages of DL pipelines to prepare high-quality datasets.

Machine learning (ML) applications within diagnostic histopathology have been extremely successful. While many successful models have been built using general-purpose models trained largely on everyday objects, there is a recent trend toward pathology-specific foundation models, trained using histopathology images. Pathology foundation models show strong performance on cancer detection and subtyping, grading, and predicting molecular diagnoses. However, we have noticed lacunae in the testing of foundation models. Nearly all the benchmarks used to test them are focused on cancer. Neoplasia is an important pathologic mechanism and key concern in much of clinical pathology, but it represents one of many pathologic bases of disease. Non-neoplastic pathology dominates findings in the placenta, a critical organ in human development, as well as a specimen commonly encountered in clinical practice. Very little to none of the data used in training pathology foundation models is placenta. Thus, placental pathology is doubly out of distribution, representing a useful challenge for foundation models. We developed benchmarks for estimation of gestational age, classifying normal tissue, identifying inflammation in the umbilical cord and membranes, and in classification of macroscopic lesions including villous infarction, intervillous thrombus, and perivillous fibrin deposition. We tested 5 pathology foundation models and 4 non-pathology models for each benchmark in tasks including zero-shot K-nearest neighbor classification and regression, content-based image retrieval, supervised regression, and whole-slide attention-based multiple instance learning. In each task, the best performing model was a pathology foundation model. However, the gap between pathology and non-pathology models was diminished in tasks related to inflammation or those in which a supervised task was performed using model embeddings. Performance was comparable among pathology foundation models. Among non-pathology models, ResNet consistently performed worse, while models from the present decade showed better performance. Future work could examine the impact of incorporating placental data into foundation model training.

The diagnosis of disease often requires analysis of a biopsy. Many diagnoses depend not only on the presence of certain features but on their location within the tissue. Recently, a number of deep learning diagnostic aids have been developed to classify digitized biopsy slides. Clinical workflows often involve processing of more than 500 slides per day. But, clinical use of deep learning diagnostic aids would require a preprocessing workflow that is cost-effective, flexible, scalable, rapid, interpretable, and transparent. Here, we present such a workflow, optimized using Dask and mixed precision training via APEX, capable of handling any patch-level or slide level classification and prediction problem. The workflow uses a flexible and fast preprocessing and deep learning analytics pipeline, incorporates model interpretation and has a highly storage-efficient audit trail. We demonstrate the utility of this package on the analysis of a prototypical anatomic pathology specimen, liver biopsies for evaluation of hepatitis from a prospective cohort. The preliminary data indicate that PathFlowAI may become a cost-effective and time-efficient tool for clinical use of Artificial Intelligence (AI) algorithms.

The inspection of stained tissue slides by pathologists is essential for the early detection, diagnosis and monitoring of disease. Recently, deep learning methods for the analysis of whole-slide images (WSIs) have shown excellent performance on these tasks, and have the potential to substantially reduce the workload of pathologists. However, successful implementation of deep learning for WSI analysis is complex and requires careful consideration of model hyperparameters, slide and image artefacts, and data augmentation. Here we introduce PathML, a Python library for performing preand post-processing of WSIs, which has been designed to interact with the most widely used deep learning libraries, PyTorch and TensorFlow, thus allowing seamless integration into deep learning workflows. We present the current best practices in deep learning for WSI analysis, and give a step-by-step guide using the PathML framework: from annotating and pre-processing of slides, to implementing neural network architectures, to training and post-processing. PathML provides a unified framework in which deep learning methods for WSI analysis can be developed and applied, thus increasing the accessibility of an important new application of deep learning.

Image classification on digital pathology images relies heavily on convolutional neural networks (CNNs), yet the behavior of alternative neural computing paragigms in this domain remains insufficiently characterized. Spiking neural networks (SNNs), which process information through event-driven spike-based dynamics, have recently become trainable at scale but have not been evaluated under standardized colorectal pathology benchmarks. This study presents the first controlled comparison of SNNs and CNNs on the Minimalist Histopathology Image Analysis (MHIST) Dataset, a widely used publicly available benchmark designed for reproducible evaluation of histopathology classification models released by Dartmouth-Hitchcock Medical Center. The classification task focuses on the clinically important binary distinction between hyperplastic polyps (HPs) and sessile serrated adenomas (SSAs), a challenging problem characterized by substantial inter-pathologist variability, where HPs are typically benign and SSAs represent precancerous lesions requiring closer clinical follow-up. Histologically, HPs exhibit superficial serrated architecture and elongated crypts, whereas SSAs are characterized by broad-based, often complex crypt structures with pronounced serration. A conventional ResNet-18 architecture and its spiking counterpart are evaluated under matched training and inference to isolate the effect of spiking computation. Models performance is quantified using the area under the receiver operating characteristic curve (ROC-AUC), yielding 0.817 for the conventional CNN and 0.812 for the SNN. This comparison enables a direct assessment of how spiking computation influences discriminative performance in HPs versus SSAs binary classification and provides a benchmark reference for SNNs on the MHIST dataset. The code is publicly available at https://github.com/qug125/snn-crcp.

This paper describes a multiclass semantic segmentation of breast cancer images into the following classes: Tumour, Stroma, Inflammatory, Necrosis and Other. The images were stained with Haematoxilin and Eosin and acquired from the Cancer Genome Atlas through the Breast Cancer Semantic Segmentation Challenge. Over 12,000 patches of data and classes were generated from training images with the use of data augmentation.The segmentation was obtained with a U-Net architecture for which the hyperparameters were explored systematically. Optimal values were obtained with batch size = 8, Loss function Adam and 50 epochs, which took over 50 hours to train. Due to this fact and limitations in time, the rest of the parameters were explored with 10 epochs and we speculate that values would increase if 50 epochs would be used. The trained U-Net was applied to unseen images, per-patch and the following metrics were obtained from full scale WSI; Accuracy, Mean Area Under the Curve and Dice Index. No post-processing was applied. The resulting segmentations outperformed the baseline in terms of accuracy for some tissues; Tumour from 0.804 to 0.913, Inflammatory from 0.743 to 0.8364. The data is available from the Grand Challenges website (https://bcsegmentation.grand-challenge.org/) and the code is available from the following GitHub repository (https://github.com/mauOrtRuiz/Breast_Cancer_Sem_Seg).

IntroductionAccurate intraoperative assessment of macrovesicular steatosis in donor liver biopsies is critical for transplantation decisions but is often limited by inter-observer variability and freezing artifacts that can obscure histological details. Artificial intelligence (AI) offers a potential solution for standardized and reproducible evaluation. To evaluate the diagnostic performance of two self-supervised learning (SSL)-based foundation models, Prov-GigaPath and UNI, for classifying macrovesicular steatosis in frozen liver biopsy sections, compared with assessments by surgical pathologists. MethodsWe retrospectively analyzed 131 frozen liver biopsy specimens from 68 donors collected between November 2022 and September 2024. Slides were digitized into whole-slide images, tiled into patches, and used to extract embeddings with Prov-GigaPath and UNI; slide-level classifiers were then trained and tested. Intraoperative diagnoses by on-call surgical pathologists were compared with ground truth determined from independent reviews of permanent sections by two liver pathologists. Accuracy was evaluated for both five-category classification and a clinically significant binary threshold (<30% vs. [&ge;]30%). ResultsFor binary classification, Prov-GigaPath achieved 96.4% accuracy, UNI 85.7%, and surgical pathologists 84.0% (P = .22). In five-category classification, accuracies were lower: Prov-GigaPath 57.1%, UNI 50.0%, and pathologists 58.7% (P = .70). Misclassification primarily occurred in intermediate categories (5%-<30% steatosis). ConclusionsSSL-based foundation models performed comparably to surgical pathologists in classifying macrovesicular steatosis, at the clinically relevant <30% vs. [&ge;]30% threshold. These findings support the potential role of AI in standardizing intraoperative evaluation of donor liver biopsies; however, the small sample size limits generalizability and requires validation in larger, balanced cohorts.

The detection of mitotic figures in histological tumor images plays a vital role in the decision-making of the appropriate therapy. However, tissue preparation and image acquisition methods degrade the performances of the deep learning-based approaches for mitotic figures detection. MItosis DOmain Generalization challenge addresses the domain-shift problem of this detection task. This work presents our approach based on preprocessing homogenizers to tackling this problem.

Computational analysis of histopathological specimens holds promise in identifying biomarkers, elucidating disease mechanisms, and streamlining clinical diagnosis. However, the application of deep learning techniques in vascular pathology remains underexplored. Here, we present a comprehensive evaluation of deep learning-based approaches to analyze digital whole-slide images of abdominal aortic aneurysm samples from 369 patients from three European centers. Deep learning demonstrated robust performance in predicting inflammatory characteristics, particularly in the adventitia, as well as fibrosis grade and remaining elastic fibers in the tunica media. Overall, this study represents the first comprehensive evaluation of computational pathology in vascular disease and has the potential to contribute to improved understanding of abdominal aortic aneurysm pathophysiology and personalization of treatment strategies, particularly when integrated with radiological phenotypes and clinical outcomes.

AO_SCPLOWBSTRACTC_SCPLOWIn many settings in digital pathology or radiology, it is of predominant importance to train classifiers that can segment disease-associated regions in medical images. While numerous deep learning approaches, most notably U-Nets, exist to learn segmentations, these approaches typically require reference segmentations as training data. As a consequence, obtaining pixel level annotations of histopathological samples has become a major bottleneck to establish segmentation learning approaches. Our contribution introduces a neural network approach to avoid the annotation bottleneck in the first place: our approach requires two-class labels such as cancer vs. healthy at the sample level only. Using these sample-labels, a meta-network is trained that infers a segmenting neural network which will segment the disease-associated region (e.g. tumor) that is present in the cancer samples, but not in the healthy samples. This process results in a network, e.g. a U-Net, that can segment tumor regions in arbitrary further samples of the same type. We establish and validate our approach in the context of digital label-free pathology, where hyperspectral infrared microscopy is used to segment and characterize the disease status of histopathological samples. Trained on a data set comprising infrared microscopic images of 100 tissue microarray spots labelled as either cancerous or cancer-free, the approach yields a U-Net that reliably identifies tumor regions or the absence of tumor in an independent test set involving 40 samples. While our present work is focused on training a U-Net for infrared microscopic images, the approach is generic in the sense that it can be adapted to other image modalities and essentially arbitrary segmenting network topologies.

Histopathology is considered the reference standard for diagnosing the presence and nature of many malignancies, including cancer. However, analyzing tissue samples and writing pathology reports is time-consuming, labor-intensive, and non-standardized. To address this problem, we present HistoGPT, the first vision language model that simultaneously generates reports from multiple pathology images. It was trained on more than 15,000 whole slide images from over 6,000 dermatology patients with corresponding pathology reports. The generated reports match the quality of human-written reports, as confirmed by a variety of natural language processing metrics and domain expert evaluations. We show that HistoGPT generalizes to six geographically diverse cohorts and can predict tumor subtypes and tumor thickness in a zero-shot fashion. Our model demonstrates the potential of an AI assistant that supports pathologists in evaluating, reporting, and understanding routine dermatopathology cases.

Spatial quantification is a critical step in most computational pathology tasks, from guiding pathologists to areas of clinical interest to discovering tissue phenotypes behind novel biomarkers. To circumvent the need for manual annotations, modern computational pathology methods have favoured multiple-instance learning approaches that can accurately predict whole-slide image labels, albeit at the expense of losing their spatial awareness. We prove mathematically that a model using instance-level aggregation could achieve superior spatial quantification without compromising on whole-slide image prediction performance. We then introduce a superpatch-based measurable multiple instance learning method, SMMILe, and evaluate it across 6 cancer types, 3 highly diverse classification tasks, and 8 datasets involving 3,850 whole-slide images. We benchmark SMMILe against 9 existing methods, and show that in all cases SMMILe matches or exceeds state-of-the-art whole-slide image classification performance while simultaneously achieving outstanding spatial quantification.

Translocations involving FN1 have been described in a variety of neoplasms, which share the presence of cartilage matrix and a variable extent of calcification. Fusions of FN1 to FGFR1 or FGFR2 have been reported in nine soft tissue chondromas, mostly demonstrated indirectly by FISH analysis. Delineation of FN1 fusions with various partner genes will facilitate our understanding of the pathogenesis and diagnostic classification of these neoplasms. In this study, we present molecular, clinical and pathologic features of 9 cartilaginous soft tissue neoplasms showing a predilection for the TMJ region and the extremities. We analyzed for gene fusions with precise breakpoints using targeted RNA-seq with a 115-gene panel, including FN1, FGFR1 and FGFR2. All 9 cases were positive for a gene fusion, including two novel fusions, FN1-MERTK and FN1-TEK, each in one case, recurrent FN1-FGFR2 in 5 cases, FN1-FGFR1 without the Ig3 domain in one case, and FGFR1-PLAG1 in one case. The breakpoints in the 5 partner gene FN1 ranged from exons 11-48, retaining the domains of signal peptide, FN1, FN2, and/or FN3, while the 3partner genes retained the trans-membrane domain, tyrosine kinase domains and /or Ig domain. The tumors with FN1-FGFR1, FN1-FGFR2 and FN1-MERTK fusions are generally characterized by nodular/lobular growth of polygonal to stellate cells within a chondroid matrix, often accompanied by various patterns of calcification. These features resemble those as described for the chondroblastoma-like variant of soft tissue chondroma. Additional histologic findings include calcium pyrophosphate dehydrate deposition and features resembling tenosynovial giant cell tumor. Overall, while the tumors from our series show significant morphologic overlap with chondroblastoma-like soft tissue chondroma, we describe novel findings that expand the morphologic spectrum of these neoplasms and have therefore labeled them as "calcified chondroid mesenchymal neoplasms." These neoplasms represent a distinct pathologic entity given the presence of recurrent FN1-receptor tyrosine kinase fusions.