Journal of Pathology Informatics最新文献

{"title":"ViCE: An automated and quantitative program to assess intestinal tissue morphology","authors":"Jeffrey La ,&nbsp;Krishnan Raghunathan ,&nbsp;Jocelyn A. Silvester ,&nbsp;Jay R. Thiagarajah","doi":"10.1016/j.jpi.2024.100397","DOIUrl":"10.1016/j.jpi.2024.100397","url":null,"abstract":"<div><h3>Background and objective</h3><div>The tissue morphology of the intestinal surface is architecturally complex with finger-like projections called villi, and glandular structures called crypts. The ratio of villus height-to-crypt depth ratio (Vh:Cd) is used to quantitatively assess disease severity and response to therapy for intestinal enteropathies, such as celiac disease and is currently quantified manually. Given the time required, manual Vh:Cd measurements have largely been limited to clinical trials and are not used widely in clinical practice. We developed ViCE (Villus Crypt Evaluator), a user-friendly software that automatically quantifies histological parameters in standard hematoxylin and eosin-stained intestinal biopsies.</div></div><div><h3>Methods</h3><div>ViCE is based on mathematical morphology operations and is scale and staining agnostic. It evaluates tissue orientation, identifies geometrical structure, and outputs key tissue measurements.</div></div><div><h3>Results</h3><div>The output measurements of Vh:Cd are concordant with manual quantifications across multiple datasets.</div></div><div><h3>Conclusions</h3><div>The underlying mathematical morphological approach for ViCE is robust, and reproducible and easily adaptable for measurement of morphological features in other tissues.</div></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100397"},"PeriodicalIF":0.0,"publicationDate":"2024-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142422215","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep feature batch correction using ComBat for machine learning applications in computational pathology","authors":"Pierre Murchan ,&nbsp;Pilib Ó Broin ,&nbsp;Anne-Marie Baird ,&nbsp;Orla Sheils ,&nbsp;Stephen P Finn","doi":"10.1016/j.jpi.2024.100396","DOIUrl":"10.1016/j.jpi.2024.100396","url":null,"abstract":"<div><h3>Background</h3><div>Developing artificial intelligence (AI) models for digital pathology requires large datasets from multiple sources. However, without careful implementation, AI models risk learning confounding site-specific features in datasets instead of clinically relevant information, leading to overestimated performance, poor generalizability to real-world data, and potential misdiagnosis.</div></div><div><h3>Methods</h3><div>Whole-slide images (WSIs) from The Cancer Genome Atlas (TCGA) colon (COAD), and stomach adenocarcinoma datasets were selected for inclusion in this study. Patch embeddings were obtained using three feature extraction models, followed by ComBat harmonization. Attention-based multiple instance learning models were trained to predict tissue-source site (TSS), as well as clinical and genetic attributes, using raw, Macenko normalized, and Combat-harmonized patch embeddings.</div></div><div><h3>Results</h3><div>TSS prediction achieved high accuracy (AUROC &gt; 0.95) with all three feature extraction models. ComBat harmonization significantly reduced the AUROC for TSS prediction, with mean AUROCs dropping to approximately 0.5 for most models, indicating successful mitigation of batch effects (e.g., CCL-ResNet50 in TCGA-COAD: Pre-ComBat AUROC = 0.960, Post-ComBat AUROC = 0.506, <em>p &lt;</em> 0.001). Clinical attributes associated with TSS, such as race and treatment response, showed decreased predictability post-harmonization. Notably, the prediction of genetic features like MSI status remained robust after harmonization (e.g., MSI in TCGA-COAD: Pre-ComBat AUROC = 0.667, Post-ComBat AUROC = 0.669, <em>p</em>=0.952), indicating the preservation of true histological signals.</div></div><div><h3>Conclusion</h3><div>ComBat harmonization of deep learning-derived histology features effectively reduces the risk of AI models learning confounding features in WSIs, ensuring more reliable performance estimates. This approach is promising for the integration of large-scale digital pathology datasets.</div></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100396"},"PeriodicalIF":0.0,"publicationDate":"2024-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142323561","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"LVI-PathNet: Segmentation-classification pipeline for detection of lymphovascular invasion in whole slide images of lung adenocarcinoma","authors":"Anna Timakova ,&nbsp;Vladislav Ananev ,&nbsp;Alexey Fayzullin ,&nbsp;Egor Zemnuhov ,&nbsp;Egor Rumyantsev ,&nbsp;Andrey Zharov ,&nbsp;Nicolay Zharkov ,&nbsp;Varvara Zotova ,&nbsp;Elena Shchelokova ,&nbsp;Tatiana Demura ,&nbsp;Peter Timashev ,&nbsp;Vladimir Makarov","doi":"10.1016/j.jpi.2024.100395","DOIUrl":"10.1016/j.jpi.2024.100395","url":null,"abstract":"<div><p>Lymphovascular invasion (LVI) in lung cancer is a significant prognostic factor that influences treatment and outcomes, yet its reliable detection is challenging due to interobserver variability. This study aims to develop a deep learning model for LVI detection using whole slide images (WSIs) and evaluate its effectiveness within a pathologist's information system. Experienced pathologists annotated blood vessels and invading tumor cells in 162 WSIs of non-mucinous lung adenocarcinoma sourced from two external and one internal datasets. Two models were trained to segment vessels and identify images with LVI features. DeepLabV3+ model achieved an Intersection-over-Union of 0.8840 and an area under the receiver operating characteristic curve (AUC-ROC) of 0.9869 in vessel segmentation. For LVI classification, the ensemble model achieved a F1-score of 0.9683 and an AUC-ROC of 0.9987. The model demonstrated robustness and was unaffected by variations in staining and image quality. The pilot study showed that pathologists' evaluation time for LVI detecting decreased by an average of 16.95%, and by 21.5% in “hard cases”. The model facilitated consistent diagnostic assessments, suggesting potential for broader applications in detecting pathological changes in blood vessels and other lung pathologies.</p></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100395"},"PeriodicalIF":0.0,"publicationDate":"2024-08-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2153353924000348/pdfft?md5=9a1e9217891b1539c144069b2cb2703f&pid=1-s2.0-S2153353924000348-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142238092","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Globalization of a telepathology network with artificial intelligence applications in Colombia: The GLORIA program study protocol","authors":"Andrés Mosquera-Zamudio ,&nbsp;Marcela Gomez-Suarez ,&nbsp;John Sprockel ,&nbsp;Julian Camilo Riaño-Moreno ,&nbsp;Emiel A.M. Janssen ,&nbsp;Liron Pantanowitz ,&nbsp;Rafael Parra-Medina","doi":"10.1016/j.jpi.2024.100394","DOIUrl":"10.1016/j.jpi.2024.100394","url":null,"abstract":"<div><p>In Colombia, cancer is recognized as a high-cost pathology by the national government and the Colombian High-Cost Disease Fund. As of 2020, the situation is most critical for adult cancer patients, particularly those under public healthcare and residing in remote regions of the country. The highest lag time for a diagnosis was observed for cervical cancer (79.13 days), followed by prostate (77.30 days), and breast cancer (70.25 days). Timely and accurate histopathological reporting plays a vital role in the diagnosis of cancer. In recent years, digital pathology has been globally implemented as a technological tool in two main areas: telepathology (TP) and computational pathology. TP has been shown to improve rapid and timely diagnosis in anatomic pathology by facilitating interaction between general laboratories and specialized pathologists worldwide through information and telecommunication technologies. Computational pathology provides diagnostic and prognostic assistance based on histopathological patterns, molecular, and clinical information, aiding pathologists in making more accurate diagnoses. We present the study protocol of the GLORIA digital pathology network, a pioneering initiative, and national grant-approved program aiming to design and pilot a Colombian digital pathology transformation focused on TP and computational pathology, in response to the general needs of pathology laboratories for diagnosing complex malignant tumors. The study protocol describes the design of a TP network to expand oncopathology services across all Colombian regions. It also describes an artificial intelligence proposal for lung cancer, one of Colombia's most prevalent cancers, and a freely accessible national histopathological image database to facilitate image analysis studies.</p></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100394"},"PeriodicalIF":0.0,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2153353924000336/pdfft?md5=861e86fc08dee64d7bef49370be8286b&pid=1-s2.0-S2153353924000336-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142075800","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Non-diagnostic time in digital pathology: An empirical study over 10 years","authors":"Aleksandar Vodovnik","doi":"10.1016/j.jpi.2024.100393","DOIUrl":"10.1016/j.jpi.2024.100393","url":null,"abstract":"","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100393"},"PeriodicalIF":0.0,"publicationDate":"2024-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2153353924000324/pdfft?md5=215132a8d517d7691de823ffcf6bf232&pid=1-s2.0-S2153353924000324-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141963948","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Engineered feature embeddings meet deep learning: A novel strategy to improve bone marrow cell classification and model transparency","authors":"Jonathan Tarquino ,&nbsp;Jhonathan Rodríguez ,&nbsp;David Becerra ,&nbsp;Lucia Roa-Peña ,&nbsp;Eduardo Romero","doi":"10.1016/j.jpi.2024.100390","DOIUrl":"10.1016/j.jpi.2024.100390","url":null,"abstract":"<div><p>Cytomorphology evaluation of bone marrow cell is the initial step to diagnose different hematological diseases. This assessment is still manually performed by trained specialists, who may be a bottleneck within the clinical process. Deep learning algorithms are a promising approach to automate this bone marrow cell evaluation. These artificial intelligence models have focused on limited cell subtypes, mainly associated to a particular disease, and are frequently presented as black boxes. The herein introduced strategy presents an engineered feature representation, the region-attention embedding, which improves the deep learning classification performance of a cytomorphology with 21 bone marrow cell subtypes. This embedding is built upon a specific organization of cytology features within a squared matrix by distributing them after pre-segmented cell regions, i.e., cytoplasm, nucleus, and whole-cell. This novel cell image representation, aimed to preserve spatial/regional relations, is used as input of the network. Combination of region-attention embedding and deep learning networks (Xception and ResNet50) provides local relevance associated to image regions, adding up interpretable information to the prediction. Additionally, this approach is evaluated in a public database with the largest number of cell subtypes (21) by a thorough evaluation scheme with three iterations of a 3-fold cross-validation, performed in 80% of the images (<em>n</em> = 89,484), and a testing process in an unseen set of images composed by the remaining 20% of the images (<em>n</em> = 22,371). This evaluation process demonstrates the introduced strategy outperforms previously published approaches in an equivalent validation set, with a f1-score of 0.82, and presented competitive results on the unseen data partition with a f1-score of 0.56.</p></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100390"},"PeriodicalIF":0.0,"publicationDate":"2024-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2153353924000294/pdfft?md5=87a5b2e97447248282a9f8d40bb281e3&pid=1-s2.0-S2153353924000294-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141629985","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Validation of AI-assisted ThinPrep® Pap test screening using the GeniusTM Digital Diagnostics System","authors":"Richard L. Cantley ,&nbsp;Xin Jing ,&nbsp;Brian Smola ,&nbsp;Wei Hao ,&nbsp;Sarah Harrington ,&nbsp;Liron Pantanowitz","doi":"10.1016/j.jpi.2024.100391","DOIUrl":"10.1016/j.jpi.2024.100391","url":null,"abstract":"<div><p>Advances in whole-slide imaging and artificial intelligence present opportunities for improvement in Pap test screening. To date, there have been limited studies published regarding how best to validate newer AI-based digital systems for screening Pap tests in clinical practice. In this study, we validated the Genius™ Digital Diagnostics System (Hologic) by comparing the performance to traditional manual light microscopic diagnosis of ThinPrep<strong>®</strong> Pap test slides. A total of 319 ThinPrep<strong>®</strong> Pap test cases were prospectively assessed by six cytologists and three cytopathologists by light microscopy and digital evaluation and the results compared to the original ground truth Pap test diagnosis. Concordance with the original diagnosis was significantly different by digital and manual light microscopy review when comparing across: (i) exact Bethesda System diagnostic categories (62.1% vs 55.8%, respectively, <em>p</em> = 0.014), (ii) condensed diagnostic categories (76.8% vs 71.5%, respectively, <em>p</em> = 0.027), and (iii) condensed diagnoses based on clinical management (71.5% vs 65.2%, respectively, <em>p</em> = 0.017). Time to evaluate cases was shorter for digital (M = 3.2 min, SD = 2.2) compared to manual (M = 5.9 min, SD = 3.1) review (t(352) = 19.44, <em>p</em> &lt; 0.001, Cohen's d = 1.035, 95% CI [0.905, 1.164]). Not only did our validation study demonstrate that AI-based digital Pap test evaluation had improved diagnostic accuracy and reduced screening time compared to light microscopy, but that participants reported a positive experience using this system.</p></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100391"},"PeriodicalIF":0.0,"publicationDate":"2024-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2153353924000300/pdfft?md5=f678b76ba4ddf0bb5fbfba56b65df94c&pid=1-s2.0-S2153353924000300-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141639228","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An explainable AI-based blood cell classification using optimized convolutional neural network","authors":"Oahidul Islam ,&nbsp;Md Assaduzzaman ,&nbsp;Md Zahid Hasan","doi":"10.1016/j.jpi.2024.100389","DOIUrl":"10.1016/j.jpi.2024.100389","url":null,"abstract":"<div><p>White blood cells (WBCs) are a vital component of the immune system. The efficient and precise classification of WBCs is crucial for medical professionals to diagnose diseases accurately. This study presents an enhanced convolutional neural network (CNN) for detecting blood cells with the help of various image pre-processing techniques. Various image pre-processing techniques, such as padding, thresholding, erosion, dilation, and masking, are utilized to minimize noise and improve feature enhancement. Additionally, performance is further enhanced by experimenting with various architectural structures and hyperparameters to optimize the proposed model. A comparative evaluation is conducted to compare the performance of the proposed model with three transfer learning models, including Inception V3, MobileNetV2, and DenseNet201.The results indicate that the proposed model outperforms existing models, achieving a testing accuracy of 99.12%, precision of 99%, and F1-score of 99%. In addition, We utilized SHAP (Shapley Additive explanations) and LIME (Local Interpretable Model-agnostic Explanations) techniques in our study to improve the interpretability of the proposed model, providing valuable insights into how the model makes decisions. Furthermore, the proposed model has been further explained using the Grad-CAM and Grad-CAM++ techniques, which is a class-discriminative localization approach, to improve trust and transparency. Grad-CAM++ performed slightly better than Grad-CAM in identifying the predicted area's location. Finally, the most efficient model has been integrated into an end-to-end (E2E) system, accessible through both web and Android platforms for medical professionals to classify blood cell.</p></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100389"},"PeriodicalIF":0.0,"publicationDate":"2024-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2153353924000282/pdfft?md5=357d6d2314681f04709e94998615c5a1&pid=1-s2.0-S2153353924000282-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141708134","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards interactive AI-authoring with prototypical few-shot classifiers in histopathology","authors":"Petr Kuritcyn ,&nbsp;Rosalie Kletzander ,&nbsp;Sophia Eisenberg ,&nbsp;Thomas Wittenberg ,&nbsp;Volker Bruns ,&nbsp;Katja Evert ,&nbsp;Felix Keil ,&nbsp;Paul K. Ziegler ,&nbsp;Katrin Bankov ,&nbsp;Peter Wild ,&nbsp;Markus Eckstein ,&nbsp;Arndt Hartmann ,&nbsp;Carol I. Geppert ,&nbsp;Michaela Benz","doi":"10.1016/j.jpi.2024.100388","DOIUrl":"10.1016/j.jpi.2024.100388","url":null,"abstract":"<div><p>A vast multitude of tasks in histopathology could potentially benefit from the support of artificial intelligence (AI). Many examples have been shown in the literature and first commercial products with FDA or CE-IVDR clearance are available. However, two key challenges remain: (1) a scarcity of thoroughly annotated images, respectively the laboriousness of this task, and (2) the creation of robust models that can cope with the data heterogeneity in the field (domain generalization). In this work, we investigate how the combination of prototypical few-shot classification models and data augmentation can address both of these challenges. Based on annotated data sets that include multiple centers, multiple scanners, and two tumor entities, we examine the robustness and the adaptability of few-shot classifiers in multiple scenarios. We demonstrate that data from one scanner and one site are sufficient to train robust few-shot classification models by applying domain-specific data augmentation. The models achieved classification performance of around 90% on a multiscanner and multicenter database, which is on par with the accuracy achieved on the primary single-center single-scanner data. Various convolutional neural network (CNN) architectures can be used for feature extraction in the few-shot model. A comparison of nine state-of-the-art architectures yielded that EfficientNet B0 provides the best trade-off between accuracy and inference time. The classification of prototypical few-shot models directly relies on class prototypes derived from example images of each class. Therefore, we investigated the influence of prototypes originating from images from different scanners and evaluated their performance also on the multiscanner database. Again, our few-shot model showed a stable performance with an average absolute deviation in accuracy compared to the primary prototypes of 1.8% points. Finally, we examined the adaptability to a new tumor entity: classification of tissue sections containing urothelial carcinoma into normal, tumor, and necrotic regions. Only three annotations per subclass (e.g., muscle and adipose tissue are subclasses of normal tissue) were provided to adapt the few-shot model, which obtained an overall accuracy of 93.6%. These results demonstrate that prototypical few-shot classification is an ideal technology for realizing an interactive AI authoring system as it only requires few annotations and can be adapted to new tasks without involving retraining of the underlying feature extraction CNN, which would in turn require a selection of hyper-parameters based on data science expert knowledge. Similarly, it can be regarded as a guided annotation system. To this end, we realized a workflow and user interface that targets non-technical users.</p></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100388"},"PeriodicalIF":0.0,"publicationDate":"2024-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2153353924000270/pdfft?md5=05adcd36f07ac4f905fe1929289c6160&pid=1-s2.0-S2153353924000270-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141415124","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Masked pre-training of transformers for histology image analysis","authors":"Shuai Jiang ,&nbsp;Liesbeth Hondelink ,&nbsp;Arief A. Suriawinata ,&nbsp;Saeed Hassanpour","doi":"10.1016/j.jpi.2024.100386","DOIUrl":"https://doi.org/10.1016/j.jpi.2024.100386","url":null,"abstract":"<div><p>In digital pathology, whole-slide images (WSIs) are widely used for applications such as cancer diagnosis and prognosis prediction. Vision transformer (ViT) models have recently emerged as a promising method for encoding large regions of WSIs while preserving spatial relationships among patches. However, due to the large number of model parameters and limited labeled data, applying transformer models to WSIs remains challenging. In this study, we propose a pretext task to train the transformer model in a self-supervised manner. Our model, MaskHIT, uses the transformer output to reconstruct masked patches, measured by contrastive loss. We pre-trained MaskHIT model using over 7000 WSIs from TCGA and extensively evaluated its performance in multiple experiments, covering survival prediction, cancer subtype classification, and grade prediction tasks. Our experiments demonstrate that the pre-training procedure enables context-aware understanding of WSIs, facilitates the learning of representative histological features based on patch positions and visual patterns, and is essential for the ViT model to achieve optimal results on WSI-level tasks. The pre-trained MaskHIT surpasses various multiple instance learning approaches by 3% and 2% on survival prediction and cancer subtype classification tasks, and also outperforms recent state-of-the-art transformer-based methods. Finally, a comparison between the attention maps generated by the MaskHIT model with pathologist's annotations indicates that the model can accurately identify clinically relevant histological structures on the whole slide for each task.</p></div>","PeriodicalId":37769,"journal":{"name":"Journal of Pathology Informatics","volume":"15 ","pages":"Article 100386"},"PeriodicalIF":0.0,"publicationDate":"2024-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2153353924000257/pdfft?md5=3dfddd9f11d8384fd0c39d65dbfab6b4&pid=1-s2.0-S2153353924000257-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141434521","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}