International Journal of Imaging Systems and Technology最新文献

Breast cancer (BC) accounts for most cancer-related deaths worldwide, so it is crucial to consider it as a prominent issue and emphasize proper diagnosis and timely detection. This study introduces a deep learning strategy called EMViT-BCC for the BC histopathology image classification to two class and eight class. The proposed model utilizes the Mobile Vision Transformer (MobileViT) block, which captures local and global features and extracts necessary features for the classification task. The proposed approach is trained and evaluated on the standard BreaKHis dataset. The model is evaluated with both the original raw histopathology images as well as the stain-normalized images for the analysis of the classification task. Extensive experiments demonstrate that the proposed EMViT-BCC achieves higher accuracy and robustness in classifying benign and malignant images and identifying various subtypes of BC. Our results demonstrate that by incorporating further layers, the classification performance of MobileViT can be greatly enhanced, with 99.43% for two-class and 93.61% for eight-class classification. These findings suggest that while stain normalization can standardize variations, original image data retain crucial details that enhance model performance. In comparison with the existing works, the proposed methodology surpasses the state-of-the-art (SOTA) methods for BC histopathology image classification. The proposed approach offers a promising solution for reliable BC classification for both binary and multi-class.

Brain tumors involve abnormal cell growth within or adjacent to brain tissues, necessitating precise segmentation for effective clinical decision-making. Traditional models often face challenges in accurately delineating tumor regions, and building robust segmentation models for high-resolution MRI data requires substantial computational power. This study presents a three-dimensional U-Net architecture with Squeeze and Excitation (SE) modules, called SE-3D Brain Net, to enhance multi-region brain tumor segmentation. The model leverages SE modules to recalibrate channel-wise feature significance, improving segmentation accuracy across tumor subregions. Extensive experiments on datasets such as BraTS 2018 and BraTS 2020 demonstrate that the model outperforms traditional U-Net models and various advanced methods, achieving average Dice scores of 0.86 for enhancing tumor, 0.84 for tumor core, and 0.86 for whole tumor segmentation. An ablation study further revealed the model's sensitivity to hyperparameters, identifying optimal settings for batch size, learning rate, and dropout rate. This study demonstrates the effectiveness of deep learning in accurately identifying brain tumors, emphasizing its potential to improve medical image analysis and patient outcomes significantly.

Automatic brain tumor segmentation technology plays a crucial role in tumor diagnosis, particularly in the precise delineation of tumor subregions. It can assist doctors in accurately assessing the type and location of brain tumors, potentially saving patients' lives. However, the highly variable size and shape of brain tumors, along with their similarity to healthy tissue, pose significant challenges in the segmentation of multi-label brain tumor subregions. This paper proposes a network model, KIDBA-Net, based on an encoder-decoder architecture, aimed at solving the issue of pixel-level classification errors in multi-label tumor subregions. The proposed Kernel Inception Depthwise Block (KIDB) employs multi-kernel depthwise convolution to extract multi-scale features in parallel, accurately capturing the feature differences between tumor types to mitigate misclassification. To ensure the network focuses more on the lesion areas and excludes the interference of irrelevant tissues, this paper adopts Bi-Cross Attention as a skip connection hub to bridge the semantic gap between layers. Additionally, the Dynamic Feature Reconstruction Block (DFRB) exploits the complementary advantages of convolution and dynamic upsampling operators, effectively aiding the model in generating high-resolution prediction maps during the decoding phase. The proposed model surpasses other state-of-the-art brain tumor segmentation methods on the BraTS2018 and BraTS2019 datasets, particularly in the segmentation accuracy of smaller and highly overlapping tumor core (TC) and enhanced tumor (ET), achieving DSC scores of 87.8%, 82.0%, and 90.2%, 88.7%, respectively; Hausdorff distances of 2.8, 2.7 mm, and 2.7, 2.0 mm.

Skin lesion classification is crucial for early diagnosis of skin cancer. However, the task faces challenges such as limited labeled data, data imbalance, and high intra-class variability. In this paper, we propose a lightweight local–global fusion (LGF) model that leverages the advantages of RegNet for local processing and Transformer for global interaction. The LGF model consists of four stages that integrate local and global pathological information using channel attention and residual connections. Furthermore, Polyloss is employed to address the data imbalance. Extensive experiments on the ISIC2018 and ISIC2019 datasets demonstrate that LGF achieves state-of-the-art performance with 93.10% and 90.36% accuracy, respectively, without any data augmentation. The LGF model is relatively lightweight and easier to reproduce, contributing to the field by offering a satisfactory trade-off between model complexity and classification performance. The code for our model will be available at https://github.com/candiceyyy/LGF.

In modern medical imaging, although there have been advances in the application of super-resolution technology to MRI in recent years, current applications still cannot meet practical needs. For example, for MRI images under specific pathological or physiological conditions, the existing super-resolution technology still lacks effectiveness in processing noise and restoring details. And when processing images with complex organizational structures, such as white matter fiber bundles in the brain, existing super-resolution techniques often fail to accurately restore image details, resulting in structural distortion. To address these deficiencies, we propose in this study an advanced super-resolution (SR) reconstruction framework tailored specifically for magnetic resonance imaging (MRI). Our approach makes use of the Denoising Diffusion Probabilistic Model (DDPM) and CrossAttention, an advanced technique known for its ability to maintain data accuracy while making the most of available conditions, leading to high-quality image restoration. By incorporating sophisticated priors and innovative network architecture, our method significantly outperforms traditional SR techniques, particularly in preserving fine anatomical details and enhancing overall image quality. The proposed framework undergoes rigorous validation through extensive experiments on diverse MRI datasets, demonstrating its robustness and effectiveness in various scenarios. Furthermore, we provide a comprehensive analysis of the performance metrics, including structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), Normalized Mean Squared Error (NMSE), and Universal Quality Index (UQI), to underscore the superiority of our DDPM-based approach. This work not only contributes to advancing the state-of-the-art in MRI SR but also paves the way for broader applications in medical imaging and related fields.

A Generative Adversarial Network (GAN) is a machine learning model used to generate new examples that are like real data. In segmentation, it can be used to improve the generation of segmented images that get closer to ground truth ones. In a super-resolution context, the GAN solves the problem of low-resolution images as it allows increasing the resolution of images while preserving the original details. In this paper, we leverage these advantages of GANs to provide a new methodology with a pipeline of two novel GANs for accurate segmentation. The proposed pipeline is composed of a first GAN model that segments the images and a second model that applies super-resolution as post-processing on the segmented images to improve its quality. The two novel GAN architectures integrate the nested residual connections (NRCs) to improve the extraction and traffic of features. These architectures are validated on CT lung datasets to detect the infected regions for COVID-19. The experimental results prove that the suggested models with NRC implementation outperform state-of-the-art solutions in multiple metrics. It achieves a dice score of 0.77 for the segmentation of COVID-19 images using the first GAN. After applying super-resolution to the segmented images using the second GAN, the PSNR and MS-SSIM metrics increase from 19.69 and 0.8756 to 33.24 and 0.9682, respectively.

Metastatic brain tumors present significant challenges in diagnosis and treatment, contributing to high mortality rates worldwide. Magnetic resonance imaging (MRI) is a pivotal diagnostic tool for identifying and assessing these tumors. However, accurate segmentation of MRI images remains critical for effective treatment planning and prognosis determination. Traditional segmentation methods, including threshold-based algorithms, often struggle with precisely delineating tumor boundaries, especially in three-dimensional (3D) images. This article introduces a 3D segmentation framework that combines Swin Transformers and 3D U-Net architectures, leveraging the complementary strengths of these models to improve segmentation accuracy and generalizability for metastatic brain tumors. We train multiple 3D U-Net and Swin U-Net models, selecting the best-performing architectures for segmenting tumor voxels. The outputs of these networks are then combined using various strategies, such as logical operations and stacking the outputs with the original images, to guide the training of a third model. Our method employs an innovative ensemble approach, integrating these outputs into a unified prediction model to enhance performance reliability. Experimental analysis on a newly released metastasis brain tumor dataset, which to the best of our knowledge has been tested for the first time using our models, yielded an impressive accuracy of 73.47%, validating the effectiveness of the proposed architectures.

Lung cancer is one of the most common and deadly diseases worldwide. The precise diagnosis of lung cancer at an early stage holds particular significance, as it contributes to enhanced therapeutic decision-making and prognosis. Despite advancements in computed tomography (CT) scanning for the detection of pulmonary nodules, accurately assessing the diverse range of pulmonary nodules continues to pose a substantial challenge. Herein, we present an innovative approach utilizing machine learning to facilitate the accurate differentiation of pulmonary nodules. Our method relies on the reconstruction of three-dimensional (3D) lung models derived from two-dimensional (2D) CT scans. Inspired by the successful utilization of deep convolutional neural networks (DCNNs) in the realm of natural image recognition, we propose a novel technique for pulmonary nodule detection employing DCNNs. Initially, we employ an algorithm to generate 3D lung models from raw 2D CT scans, thereby providing an immersive stereoscopic depiction of the lungs. Subsequently, a DCNN is introduced to extract features from images and classify the pulmonary nodules. Based on the developed model, pulmonary nodules with various features have been successfully classified with 86% accuracy, demonstrating superior performance. We hold the belief that our strategy will provide a useful tool for the early clinical diagnosis and management of lung cancer.

Cervical cancer persists as a significant global health concern, underscoring the vital importance of early detection for effective treatment and enhanced patient outcomes. While traditional Pap smear tests remain an invaluable diagnostic tool, they are inherently time-consuming and susceptible to human error. This study introduces an innovative approach that employs convolutional neural networks (CNN) to enhance the accuracy and efficiency of cervical cell classification. The proposed model incorporates the Convolutional Block Attention Module (CBAM) and parallel branch architectures, which facilitate enhanced feature extraction by focusing on crucial spatial and channel information. The process of feature extraction entails the identification and utilization of the most pertinent elements within an image for the purpose of classification. The proposed model was meticulously assessed on the SIPaKMeD dataset, attaining an exceptional degree of accuracy (92.82%), which surpassed the performance of traditional CNN models. The incorporation of sophisticated attention mechanisms enables the model to not only accurately classify images but also facilitate interpretability by emphasizing crucial regions within the images. This study highlights the transformative potential of cutting-edge deep learning techniques in medical image analysis, particularly for cervical cancer screening, providing a powerful tool to support pathologists in early detection and accurate diagnosis. Future work will explore additional attention mechanisms and extend the application of this architecture to other medical imaging tasks, further enhancing its clinical utility and impact on patient outcomes.

Owing to legal restrictions and privacy preservation, it is impractical to consolidate medical data across multiple regions for model training, leading to difficulties in data sharing. Federated learning (FL) methods present a solution to this issue. However, traditional FL encounters difficulties in handling non-independent identically distributed (Non-IID) data, where the data distribution across clients is heterogeneous and not uniformly distributed. Although personalized federated learning (PFL) can tackle the Non-IID issue, it has drawbacks such as lower accuracy rates or high memory usage. Furthermore, knowledge-distillation-based PFL exhibits shortcomings in model learning capabilities. In this study, we propose FedSPD, a novel federated learning framework that integrates similarity-preserving knowledge distillation to bridge the gap between global knowledge and local models. FedSPD reduces discrepancies by aligning feature representations through cosine similarity at the feature level, enabling local models to assimilate global knowledge while preserving personalized characteristics. This approach enhances model performance in heterogeneous environments while mitigating privacy risks by sharing only averaged logits, in line with stringent medical data security requirements. Extensive experiments were conducted on three datasets: MNIST, CIFAR-10, and brain tumor MRI, comparing FedSPD with nine state-of-the-art FL and PFL algorithms. On general datasets, under the IID setting, FedSPD achieved performance comparable to existing methods. In Non-IID scenarios, we employed the Dirichlet distribution to control the data distribution across clients, allowing us to model and assess non-uniform data partitions in our FL settings. FedSPD demonstrated exceptional performance, with accuracy improvements of up to 77.77% over traditional FL methods and up to 4.19% over PFL methods. On the brain tumor MRI dataset, FedSPD outperformed most algorithms under the IID condition. In Non-IID settings, it exhibited even greater advantages, with accuracy improvements of up to 78.41% over traditional FL methods and up to 10.55% over PFL methods. Additionally, FedSPD significantly reduced computational overhead, shortening each training round by up to 67.25% compared to other PFL methods and reducing parameter size by up to 49.34%, thereby improving scalability and efficiency. By effectively integrating global and personalized features, FedSPD not only enhanced model generalization across heterogeneous medical datasets but also strengthened clinical decision-making, contributing to more accurate diagnoses and better patient prognosis. This scalable and privacy-preserving solution meets the practical demands of healthcare applications.