Lecture notes in networks and systems, Journal Year: 2024, Volume and Issue: unknown, P. 123 - 135
Published: Jan. 1, 2024
Language: Английский
Lecture notes in networks and systems, Journal Year: 2024, Volume and Issue: unknown, P. 123 - 135
Published: Jan. 1, 2024
Language: Английский
Diagnostics, Journal Year: 2024, Volume and Issue: 14(14), P. 1472 - 1472
Published: July 9, 2024
With the improvement of economic conditions and increase in living standards, people's attention regard to health is also continuously increasing. They are beginning place their hopes on machines, expecting artificial intelligence (AI) provide a more humanized medical environment personalized services, thus greatly expanding supply bridging gap between resource demand. development IoT technology, arrival 5G 6G communication era, enhancement computing capabilities particular, application AI-assisted healthcare have been further promoted. Currently, research field assistance deepening expanding. AI holds immense value has many potential applications institutions, patients, professionals. It ability enhance efficiency, reduce costs, improve quality intelligent service experience for professionals patients. This study elaborates history timelines field, types technologies informatics, opportunities challenges medicine. The combination profound impact human life, improving levels life changing lifestyles.
Language: Английский
Citations
28Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 95, P. 106439 - 106439
Published: May 13, 2024
The cell nuclei in pathology images provide clinicians with important tissue information for diagnosis. However, nucleus segmentation faces various challenges such as image blurring, blurred boundaries, noise, and holes. Existing methods are difficult to achieve the expected results field of nuclei. Based on this, this study proposes a framework based deblurring region proxies (DRPVit) medical decision-making systems. First, we employ an information-enhanced iterative filtering adaptive network (IE-IFANet) improve clarity regions boundaries. Then, considering performance complexity network, RegProxy model is used segmentation. utilizes agent represent homogeneous semantics, transformer encoder relationship between these regions, finally classifier prediction. Finally, introduce combined loss function consisting boundary tversky balance optimize edges regions. In post-processing stage, eliminated holes noise predicted introduced paramedical tools measure size morphological differences. experimental evaluation show that our method outperforms comparative models improvement 3.3%, 2.3%, 2.1% Intersection over Union, Dice similarity coefficient, Recall, respectively.
Language: Английский
Citations
10Biomedical & Pharmacology Journal, Journal Year: 2025, Volume and Issue: 18(December Spl Edition), P. 121 - 138
Published: Jan. 20, 2025
Digital medical images study, analysis and processing is one of visualization areas, which helps to improve diagnostic diseases detection, monitoring their progression treatment. In this regard, the cytological important effective, it allows study various cellular structures. One components such research allocation potential areas interest, considering specifics color presentation. The paper proposes a new combined approach for identifying interest based on edge detection operators, compared with corresponding classical methods. It shown that proposed gives results no worse than methods, some types images, even better. At same time, resulting quality assessments compactness achieved, range obtained making effective decisions expanded depending specification goals. semi-transparent background, low-contrast difference between background objects approach, in comparison approaches, provides superiority terms niqe assessment at least 10%, brisque more 20%, derivative (ME AE) – 7.5% 1%, respectively. details processed best all cases. an answer combination operators individual channels without using pre-processing methods original image. This increased efficiency clinical trials diagnosis.
Language: Английский
Citations
1IET Image Processing, Journal Year: 2023, Volume and Issue: 17(10), P. 3040 - 3054
Published: June 30, 2023
Abstract Due to the improvement in computing power and development of computer technology, deep learning has pene‐trated into various fields medical industry. Segmenting lesion areas scans can help clinicians make accurate diagnoses. In particular, convolutional neural networks (CNNs) are a dominant tool vision tasks. They accurately locate classify areas. However, due their inherent inductive bias, CNNs may lack an understanding long‐term dependencies images, leading less grasping details images. To address this problem, we explored Transformer‐based solution studied its feasibility imaging tasks (OstT). First, performed super‐resolution reconstruction on original MRI image osteosarcoma improved texture features tissue structure reduce error caused by unclear during model training. Then, propose method for segmentation. A gated axial attention is used, which augments existing architectures introducing additional control mechanism self‐attention module improve segmentation accuracy. Experiments real datasets show that our outper‐forms models such as Unet. It effectively assist doctors examinations.
Language: Английский
Citations
23Biomedical Signal Processing and Control, Journal Year: 2023, Volume and Issue: 89, P. 105814 - 105814
Published: Dec. 5, 2023
Language: Английский
Citations
21Complex & Intelligent Systems, Journal Year: 2024, Volume and Issue: 10(4), P. 5831 - 5849
Published: May 23, 2024
Abstract Malignant tumors are a common cytopathologic disease. Pathological tissue examination is key tool for diagnosing malignant tumors. Doctors need to manually analyze the images of pathological sections, which not only time-consuming but also highly subjective, easily leading misdiagnosis. Most existing computer-aided diagnostic techniques focus too much on accuracy when processing images, and do take into account problems insufficient resources in developing countries meet training large models difficulty obtaining medical annotation data. Based this, this study proposes an artificial intelligence multiprocessing scheme (MSPInet) digital pathology We use such as data expansion noise reduction enhance dataset. Then we design coarse segmentation method cell nuclei based Transformer Semantic Segmentation further optimize tumor edges using conditional random fields. Finally, improve strategy knowledge distillation. As assistive system, can quantify convert complex analyzable image information. Experimental results show that our performs well terms has advantages time space efficiency. This makes technology available resourced, equipped care. The teacher model lightweight student included achieve 71.6% 66.1% Intersection over Union (IoU) respectively, outperforming Swin-unet CSWin Transformer.
Language: Английский
Citations
8Egyptian Informatics Journal, Journal Year: 2024, Volume and Issue: 27, P. 100530 - 100530
Published: Aug. 28, 2024
Language: Английский
Citations
8Complex & Intelligent Systems, Journal Year: 2024, Volume and Issue: 10(3), P. 4253 - 4274
Published: March 4, 2024
Abstract Artificial intelligence has made substantial progress in many medical application scenarios. The quantity and complexity of pathology images are enormous, but conventional visual screening techniques labor-intensive, time-consuming, subject to some degree subjectivity. Complex pathological data can be converted into mineable image features using artificial analysis technology, enabling professionals quickly quantitatively identify regions interest extract information about cellular tissue. In this study, we designed a assistance system for segmenting quantifying statistical results, including enhancement, cell nucleus segmentation, model tumor, quantitative analysis. address the problem uneven healthcare resources, high-precision teacher (HRMED_T) lightweight student (HRMED_S). HRMED_T is based on Transformer high-resolution representation learning. It achieves accurate segmentation by parallel low-resolution convolution high-scaled iterative fusion, while also maintaining representation. HRMED_S Channel-wise Knowledge Distillation approach simplify structure, achieve faster convergence, refine results conditional random fields instead fully connected structures. experimental show that our better performance than other methods. Intersection over Union (IoU) reaches 0.756. IoU 0.710 params only 3.99 M.
Language: Английский
Citations
7Complex & Intelligent Systems, Journal Year: 2024, Volume and Issue: 10(5), P. 6031 - 6050
Published: May 27, 2024
Abstract Magnetic resonance imaging (MRI) examinations are a routine part of the cancer treatment process. In developing countries, disease diagnosis is often time-consuming and associated with serious prognostic problems. Moreover, MRI characterized by high noise low resolution. This creates difficulties in automatic segmentation lesion region, leading to decrease performance model. paper proposes deep convolutional neural network osteosarcoma image system based on reduction super-resolution reconstruction, which first time introduce methods task segmentation, effectively improving Model generalization performance. We refined initial dataset using Differential Activation Filter, separating those data that had little effect model training. At same time, we carry out rough denoising image. Then, an improved information multi-distillation adaptive cropping proposed reconstruct original improve resolution Finally, high-resolution used segment image, boundary optimized provide reference for doctors. Experimental results show this algorithm has stronger anti-noise ability than existing methods. Code: https://github.com/GFF1228/NSRDN.
Language: Английский
Citations
7IET Image Processing, Journal Year: 2023, Volume and Issue: 18(1), P. 175 - 193
Published: Sept. 26, 2023
Abstract Artificial intelligence decision systems play an important supporting role in the field of medical information. Medical image analysis is part and even more diagnosis treatment. The wealth cellular information histopathological images makes them a reliable means diagnosing tumors. However, due to large size, high resolution, complex background structure pathology images, deep learning methods still have various difficulties recognition images. Based on this, this study proposes two‐stage continuous improvement‐based approach for systems. For with backgrounds, normalization enhancement performed remove effects noise color, light‐dark inconsistencies segmentation network. refinement PSP Net (CRPSPNet) then designed accurate CRPSPNet divided into two stages: Pyramid Scene Parsing Network obtain coarse results; model refines results first stage. Experiments using than 1,000 osteosarcoma shown that method gives fewer computer resources processing time traditional optimization models. Its Intersection over Union achieves 0.76.
Language: Английский
Citations
14