Emerging topics in statistics and biostatistics, Год журнала: 2024, Номер unknown, С. 157 - 190
Опубликована: Янв. 1, 2024
Язык: Английский
Emerging topics in statistics and biostatistics, Год журнала: 2024, Номер unknown, С. 157 - 190
Опубликована: Янв. 1, 2024
Язык: Английский
Results in Engineering, Год журнала: 2024, Номер 24, С. 103054 - 103054
Опубликована: Окт. 8, 2024
Язык: Английский
Процитировано
5Journal of Imaging, Год журнала: 2024, Номер 10(7), С. 151 - 151
Опубликована: Июнь 21, 2024
Nonmydriatic retinal fundus images often suffer from quality issues and artifacts due to ocular or systemic comorbidities, leading potential inaccuracies in clinical diagnoses. In recent times, deep learning methods have been widely employed improve image quality. However, these require large datasets lack robustness settings. Conversely, the inherent stability adaptability of traditional unsupervised methods, coupled with their reduced reliance on extensive data, render them more suitable for real-world applications, particularly limited data context high noise levels a significant presence artifacts. existing encounter challenges such as sensitivity outliers, assumptions like cluster shapes, difficulties scalability interpretability, when utilized enhancement. To tackle challenges, we propose novel robust PCA (RPCA) method low-rank sparse decomposition that also integrates affine transformations τi, weighted nuclear norm, L2,1 norms, aiming overcome limitations achieve improvement unseen by methods. We employ norm (Lw,∗) assign weights singular values each utilize eliminate correlated samples outliers images. Moreover, τi is enhance alignment, making new variations, noise, blurring. The Alternating Direction Method Multipliers (ADMM) used optimally determine parameters, including solving an optimization problem. Each parameter addressed separately, harnessing benefits ADMM. Our introduces update approach significantly improves quality, detecting cataracts, diabetic retinopathy. Simulation results confirm our method's superiority over state-of-the-art across various datasets.
Язык: Английский
Процитировано
3Applied Sciences, Год журнала: 2025, Номер 15(10), С. 5693 - 5693
Опубликована: Май 20, 2025
Artificial intelligence-based biomedical image processing has become an important area of research in recent decades. In this context, one the most problems encountered is close contrast values between pixels to be segmented and remaining pixels. Among crucial advantages provided by metaheuristic algorithms, they are generally able provide better performances segmentation images due their randomized gradient-free global search abilities. Math-inspired algorithms can considered robust groups while also presenting non-complex structures. work, recently proposed Circle Search Algorithm (CSA), Tangent (TSA), Arithmetic Optimization (AOA), Generalized Normal Distribution (GNDO), Global Method based on Clustering Parabolic Approximation (GOBC-PA), Sine Cosine (SCA) were implemented for clustering then applied retinal vessel task from DRIVE STARE databases. Firstly, results each algorithm obtained compared with other. Then, compare statistical analyses carried out terms sensitivity (Se), specificity (Sp), accuracy (Acc), standard deviation, Wilcoxon rank-sum test results. Finally, detailed convergence speed, mean squared error (MSE), CPU time, number function evaluations (NFEs) metrics.
Язык: Английский
Процитировано
0Emerging topics in statistics and biostatistics, Год журнала: 2024, Номер unknown, С. 157 - 190
Опубликована: Янв. 1, 2024
Язык: Английский
Процитировано
1