International Journal of Imaging Systems and Technology最新文献

Accurate segmentation of brain tumors in magnetic resonance imaging (MRI) is critical for diagnosis, treatment planning, and longitudinal monitoring. The development of robust and generalizable deep learning models for tumor segmentation is hindered by challenges such as data privacy, limited annotations, and domain variability across clinical institutions. To address these issues, we propose a dual-model federated learning model for brain tumor segmentation that enables collaborative model training without sharing patient data. The model employs two specialized architectures: a Multi-Scale Encoder U-Net (MSE-UNet) for fine-grained, multi-resolution feature extraction and a Residual Attention Transpose U-Net (ART-UNet) that leverages residual learning and dual attention mechanisms to enhance contextual sensitivity and robustness under non-IID conditions. To ensure effective learning across distributed, heterogeneous clients, we introduce a Dual-Model Architecture-Aware Aggregation (DAAA) strategy, which performs independent, performance-weighted aggregation of each architecture's updates. The proposed method is evaluated on two benchmark datasets, BraTS 2018 and TCGA-LGG, demonstrating superior performance compared to several state-of-the-art baselines. The model achieves Dice scores of 91.30% and 90.10% on BraTS and TCGA-LGG, respectively, with improved IoU, sensitivity, and boundary precision. Ablation studies confirm that each component, including auxiliary supervision, architectural duality, and adaptive aggregation, contributes significantly to overall performance. Clinically, this framework offers a scalable and privacy-preserving solution that can be integrated into real-world healthcare systems without compromising patient data security. By enabling cross-institutional collaboration and ensuring robust performance across diverse imaging protocols, the proposed model facilitates early and accurate tumor delineation, supporting radiologists in critical decision-making processes such as surgical planning, radiotherapy guidance, and follow-up assessment. This study establishes a foundation for practical deployment in federated clinical environments, particularly within resource-constrained or privacy-sensitive institutions.

Skin cancer is a prevalent and potentially life-threatening condition that requires accurate and timely diagnosis. Dermoscopic imaging aids in diagnosis but is often limited by manual interpretation, which can be time–intensive and error–prone. Automated classification systems using convolutional neural networks (CNNs) have shown significant promise in enhancing accuracy and efficiency. To further improve classification performance, we propose a novel deep learning-based Multi–Scale Feature Map Fusion (MFMF) model that extracts and fuses features from multiple convolutional layers. The MFMF module effectively combines these multi–scale features, enabling robust feature capture even with limited datasets. Additionally, a hair removal algorithm is proposed to enhance prediction accuracy through improved image preprocessing. On the HAM10000 dataset, our proposed model achieves an overall accuracy of 96.12%, an AUC of 98.7%, and an F1 Score of 93.29%, along with notable improvements in precision, recall, and sensitivity.

Accurate organ segmentation from magnetic resonance imaging (MRI) or computed tomography (CT) images is essential for surgical planning and decision-making. Traditional fully supervised deep learning methods often exhibit a significant decline in performance when applied to datasets that differ from the training data, thus limiting their clinical applicability. This study proposes a novel segmentation method based on unsupervised domain adaptation, aiming to improve cross-domain segmentation performance without the need for ground truth labels in the target domain. Specifically, our method trains the network with labeled source images and unlabeled target images, introducing a bidirectional feature-prototype contrastive loss to align features across domains, minimizing within-class variations and maximizing between-class variations. To further improve model performance, we propose a prototype-guided pseudo-label fusion module that generates high-quality pseudo-labels for the unlabeled target images between domain prototypes. Experimental results show that our method outperforms other unsupervised domain adaptation segmentation approaches, achieving state-of-the-art performance. Code is available at: https://github.com/WANGSIQII/UDA.git.

Automatic segmentation and classification methods of glioblastomas in magnetic resonance imaging (MRI) scans are essential to overcome the limitations of error-prone manual methods, especially given the intrinsic challenges such as intensity nonuniformity, diverse anatomical brain alterations, and significant variations in tumor shape, size, and location. These complexities pose major hurdles for radiologists in diagnosis and surgical planning, underscoring the critical significance of robust automated solutions. This study presents a novel fully automated approach for segmentation and classification of glioblastoma brain tumors using multi-spectral MRI data. Our proposed framework innovatively integrates two key steps. In the first step, a new level set method is presented for segmentation, which is uniquely enhanced by super-pixel fuzzy entropy-based clustering—a technique designed to effectively handle image inhomogeneities and noise—density peak clustering, and a lattice Boltzmann solver for efficient contour evolution. In the second step, a VGG-16 deep neural network is employed for precise classification. To assess the capability of the proposed method in both segmentation and classification tasks, real T2-weighted and fluid-attenuated inversion recovery magnetic resonance images of glioblastomas from the BraTS 2020 dataset are used simultaneously in a multi-spectral manner. Our segmentation results, evaluated by measuring the Dice coefficient, Jaccard index, sensitivity, specificity, and running time. The mean values (Mean ± Standard deviation) of these metrics are 0.8915 ± 0.0293, 0.8055 ± 0.0478, 0.9535 ± 0.0644, 0.9910 ± 0.0364, 2.2909 ± 0.2597, respectively. Additionally, the average values of accuracy, precision, recall, and F1-score across the fivefold cross-validation of the classification method are 0.9149, 0.9532, 0.9160, and 0.9349, respectively. According to the experiments, our proposed fully automated framework not only achieves superior performance in simultaneous segmentation and classification compared to other state-of-the-art segmentation methods but also offers a robust and efficient solution for clinical applications. While this study demonstrates strong potential, future work will focus on extending the framework for multi-label segmentation of different tumor sub-regions and validating its efficacy on even larger and more diverse clinical datasets.

Diabetic retinopathy is a prevalent and sight-threatening complication of diabetes that affects individuals worldwide. Effectively addressing this condition requires adapting approaches to the specific characteristics of retinal images. Existing works often tackle the diagnostic challenge without focusing on a specific aspect. In contrast, our study introduces a new problem-oriented strategy that addresses key gaps in diabetic retinopathy using three novel, tailored approaches. First, to address the underexploitation of high-resolution retinal images, we propose a resolution-preserving, data-based approach that employs patch-based analysis without downscaling while also mitigating data scarcity and imbalance. Second, inspired by real-world clinical practice, we develop a symptoms-based approach that explicitly segments multiple key pathological indicators (blood vessels, exudates, and microaneurysms) and then uses them to guide the classification network. Third, we propose a hierarchical approach that decomposes the multi-stage classification task into multiple hierarchical binary classifications, enabling more specialized feature learning and informed decision-making across different severity levels. Evaluations on both EyePACS and APTOS benchmark datasets showcased superior performance, surpassing or matching contemporary state-of-the-art results. These outcomes demonstrate the effectiveness of our proposed approaches and underscore the strategy's potential to improve diabetic retinopathy diagnosis.

Fundus retinal vessel segmentation is important for assisting in the diagnosis and monitoring of related ophthalmic diseases. Due to the fact that fundus retinal vessels have the characteristics of both local complex topology (e.g., branching structure) and global wide-area distribution, to be able to simultaneously take into account the local detail information and global context information and fully fuse the two kinds of information, this paper proposes a feature aggregation network (FANet) with dual encoders for fundus retinal vessel segmentation. Firstly, we employ the convolutional neural network (CNN) and Transformer to construct dual path encoders for extracting local detail information and global context information, respectively. Among them, to enhance the feature expression ability of the feed-forward network (FFN) in the Transformer block, we design the feature-optimized FFN (F3N). Next, we introduce the dual path feature aggregation (DPFA) module to fully fuse the feature information extracted from the CNN and Transformer paths. Finally, we introduce the multi-scale feature aggregation (MFA) module to obtain rich multi-scale information and adapt to the scale variation of vessels. Experimental results on CHASE-DB1, DRIVE, and STARE datasets demonstrate that FANet outperforms the existing mainstream segmentation methods in the comprehensive performance comparison of multiple evaluation metrics, verifying its effectiveness.