Knowledge-Based Systems, Год журнала: 2023, Номер 275, С. 110721 - 110721
Опубликована: Июнь 15, 2023
Язык: Английский
Knowledge-Based Systems, Год журнала: 2023, Номер 275, С. 110721 - 110721
Опубликована: Июнь 15, 2023
Язык: Английский
Expert Systems with Applications, Год журнала: 2023, Номер 241, С. 122666 - 122666
Опубликована: Ноя. 23, 2023
Язык: Английский
Процитировано
131Artificial Intelligence Review, Год журнала: 2023, Номер 56(S3), С. 2917 - 2970
Опубликована: Окт. 4, 2023
Язык: Английский
Процитировано
96IEEE Journal of Biomedical and Health Informatics, Год журнала: 2024, Номер 28(6), С. 3557 - 3570
Опубликована: Март 8, 2024
Grading laryngeal squamous cell carcinoma (LSCC) based on histopathological images is a clinically significant yet challenging task. However, more low-effect background semantic information appeared in the feature maps, channels, and class activation which caused serious impact accuracy interpretability of LSCC grading. While traditional transformer block makes extensive use parameter attention, model overlearns information, resulting ineffectively reducing proportion semantics. Therefore, we propose an end-to-end network with transformers constrained by learned-parameter-free attention (LA-ViT), improve ability to learn high-effect target reduce Firstly, according generalized linear probabilistic, demonstrate that (LA) has stronger highly effective than attention. Secondly, first-type LA LA-ViT utilizes map position subspace realize query. Then, it uses channel key, adopts average convergence obtain value. And those construct mechanism. Thus, reduces semantics maps channels. Thirdly, second-type probability matrix decision level weight key query, respectively. So, maps. Finally, build new complex pathology image dataset address problem, less research grading models because lacking meaningful datasets. After experiments, whole metrics outperform other state-of-the-art methods, visualization match better regions interest pathologists' decision-making. Moreover, experimental results conducted public show superior generalization performance methods.
Язык: Английский
Процитировано
20Expert Systems with Applications, Год журнала: 2023, Номер 230, С. 120637 - 120637
Опубликована: Июнь 1, 2023
Язык: Английский
Процитировано
33BioMedical Engineering OnLine, Год журнала: 2023, Номер 22(1)
Опубликована: Сен. 25, 2023
Transformers have been widely used in many computer vision challenges and shown the capability of producing better results than convolutional neural networks (CNNs). Taking advantage capturing long-range contextual information learning more complex relations image data, applied to histopathological processing tasks. In this survey, we make an effort present a thorough analysis uses analysis, covering several topics, from newly built Transformer models unresolved challenges. To be precise, first begin by outlining fundamental principles attention mechanism included other key frameworks. Second, analyze Transformer-based applications imaging domain provide evaluation 100 research publications across different downstream tasks cover most recent innovations, including survival prediction, segmentation, classification, detection, representation. Within survey work, also compare performance CNN-based techniques based on recently published papers, highlight major challenges, interesting future directions. Despite outstanding architectures number papers reviewed anticipate that further improvements exploration are still required future. We hope paper will give readers field study understanding up-to-date list summary provided at https://github.com/S-domain/Survey-Paper .
Язык: Английский
Процитировано
28IEEE Journal of Biomedical and Health Informatics, Год журнала: 2024, Номер 28(4), С. 2091 - 2102
Опубликована: Янв. 9, 2024
Digital pathology images' extensive cellular information provide a trustworthy foundation for tumor diagnosis. With the aid of computer-aided diagnostics, pathologists can locate crucial more quickly. The cascade structure refines segmentation results by utilizing its multi-task and multi-stage characteristics. However, cascade-based models require downsampling cropping patches during inference process due to ultra-high resolution complex images. This not only increases cost computation time but also in loss details corrupts global contextual information. study proposes Pathology Image Assistance Program (CRSDPI) medical decision-making systems that is based on continuous improvement. After locating region interest using maximum inter-class variance method, pictures are preprocessed account impacts staining inconsistencies sensitivity variations model's performance. Ultimately, we create two-phase continuously refined network (TCRNet) combining an enhanced refinement model with coarse built pyramid scene parsing network. introduces auxiliary term speed up convergence, implicit function reduce computational reconstruct details. TCRNet target successively aligning features without need take cascading decoder operations after encoder. Experiments conducted digital images breast cancer osteosarcoma demonstrate superior prediction accuracy our strategy.
Язык: Английский
Процитировано
17Expert Systems with Applications, Год журнала: 2024, Номер 249, С. 123549 - 123549
Опубликована: Фев. 20, 2024
Язык: Английский
Процитировано
14Neural Computing and Applications, Год журнала: 2025, Номер unknown
Опубликована: Янв. 10, 2025
Язык: Английский
Процитировано
1Expert Systems with Applications, Год журнала: 2024, Номер 252, С. 124113 - 124113
Опубликована: Май 1, 2024
Язык: Английский
Процитировано
9Scientific Reports, Год журнала: 2024, Номер 14(1)
Опубликована: Авг. 17, 2024
The second most common type of malignant tumor worldwide is colorectal cancer. Histopathology image analysis offers crucial data for the clinical diagnosis Currently, deep learning techniques are applied to enhance cancer classification and localization in histopathological analysis. Moreover, traditional might loss integrated information while evaluating thousands patches recovered from whole slide images (WSIs). This research proposes a novel detection network (CCDNet) that combines coordinate attention transformer with atrous convolution. CCDNet first denoises input using Wiener based Midpoint weighted non-local means filter (WMW-NLM) guaranteeing precise diagnoses maintain features. Also, convolution (AConvCAT) introduced, which successfully advantages two networks classify tissue at various scales by capturing local global information. Further, model Cross-shaped window (CrSWin) tiny changes multiple angles. proposed achieved accuracy rates 98.61% 98.96%, on histological NCT-CRC-HE-100 K datasets correspondingly. comparison demonstrates suggested framework performed better than advanced methods already use. In hospitals, clinicians can use verify diagnosis.
Язык: Английский
Процитировано
9