International Journal of Imaging Systems and Technology最新文献

Early prediction and timely diagnosis of ocular diseases are of great significance for preventing vision loss and improving patients' quality of life. However, many eye diseases exhibit atypical symptoms in their early stages, leading to delays in clinical diagnosis and consequently postponing treatment opportunities, which may worsen the condition. Artificial intelligence (AI), particularly deep learning, has provided novel solutions for the automated detection and prediction of ophthalmic diseases. Methods such as convolutional neural networks (CNNs), transfer learning, generative adversarial networks (GANs), recurrent neural networks (RNNs), attention mechanisms, and interpretable deep learning have been widely applied in the analysis of fundus and optical coherence tomography (OCT) images. These approaches could extract subtle features that are difficult to capture using traditional techniques, achieving outstanding performance in the diagnosis and grading of diseases such as diabetic retinopathy, glaucoma, and macular degeneration. Therefore, a systematic analysis of the current applications and development trends of AI-based methods in ocular disease prediction, along with a discussion of their advantages, limitations, and future improvement directions in alignment with clinical needs, will provide valuable insights for advancing intelligent diagnostic research and practice in retinal diseases.

Breast cancer detection remains a diagnostic challenge, particularly in dense breast tissues. This study aimed to develop and evaluate an integrative deep learning framework that leverages 3D digital breast tomosynthesis (DBT) to improve accuracy, interpretability, and generalizability of breast cancer classification across clinical settings. This framework focuses on image-level classification of normal, benign, and malignant cases, rather than lesion-level detection, utilizing DBT images to enhance accuracy and generalizability. A total of 2589 DBT images were collected from six medical centers, categorized into normal, benign, and malignant cases. The framework incorporated both convolutional neural network (CNN)-based transfer learning models (ResNet50, InceptionV3, EfficientNet-B0) and transformer-based models (Vision Transformer, Swin Transformer). Images underwent preprocessing including normalization, filtering, resizing, and augmentation. All models were trained under a 10-fold cross-validation scheme with an external validation set retained from an independent institution. Performance was evaluated using accuracy, recall, precision, F1-score, and AUC-ROC. To enhance interpretability, Grad-CAM and t-SNE visualizations were generated. Statistical tests (paired t-test and Wilcoxon) assessed performance significance, and a radiologist independently scored Grad-CAM heatmaps to evaluate clinical plausibility. Transformer models consistently outperformed CNN architectures across all metrics. Swin Transformer achieved the highest accuracy (94%) and AUC (96%) in both internal and external validations, particularly excelling in malignant lesion detection. Grad-CAM and t-SNE demonstrated superior spatial focus and feature separability in transformer models. Statistical tests confirmed the significance of performance differences (p < 0.05), and radiologist scoring revealed strong clinical alignment in transformer heatmaps (mean score: 1.8). Computational efficiency analysis indicated faster convergence in transformer models despite higher training costs. The proposed framework offers a clinically interpretable and diagnostically accurate solution for DBT-based breast cancer detection. Transformer architectures, combined with advanced visual analytics, demonstrate potential for real-world deployment and clinical decision support.

In recent years, transformer-based methods have achieved remarkable progress in medical image segmentation due to their superior ability to capture long-range dependencies. However, these methods typically suffer from two major limitations. First, their computational complexity scales quadratically with the input sequences. Second, the feed-forward network (FFN) modules in vanilla Transformers typically rely on fully connected layers, which limits models' ability to capture local contextual information and multiscale features critical for precise semantic segmentation. To address these issues, we propose an efficient medical image segmentation network, named TCSAFormer. The proposed TCSAFormer adopts two key ideas. First, it incorporates a Compressed Attention (CA) module, which combines token compression and pixel-level sparse attention to dynamically focus on the most relevant key-value pairs for each query. This is achieved by pruning globally irrelevant tokens and merging redundant ones, significantly reducing computational complexity while enhancing the model's ability to capture relationships between tokens. Second, it introduces a Dual-Branch Feed-Forward Network (DBFFN) module as a replacement for the standard FFN to capture local contextual features and multiscale information, thereby strengthening the model's feature representation capability. We conduct extensive experiments on four publicly available medical image segmentation datasets: ISIC-2018, CVC-ClinicDB, Synapse and Abdomen MRI, to evaluate the segmentation performance of TCSAFormer. Experimental results demonstrate that TCSAFormer achieves superior performance compared to existing state-of-the-art (SOTA) methods, while maintaining lower computational overhead, thus achieving an optimal trade-off between efficiency and accuracy. The code is available on GitHub.

Glaucoma is a dangerous disease that can lead to blindness in advanced stages. It has been a hot topic among machine learning researchers as it can be diagnosed quickly and effectively through optic disc segmentation. However, anomalies in the optic disk region significantly complicate this task. There is also a need for a model that produces stable results on different data sets. Therefore, in this study, we propose a novel method that can be used for early diagnosis and treatment of glaucoma. Using deep learning architecture, hypercolumn deep features extracted from retinal fundus images were trained with different classifiers, and their behavior was examined in depth according to varying thresholding values. Global and local thresholding approaches were developed to improve the predictions of the standard classifier. The proposed model achieved Dice scores of 0.9437, 0.9654, and 0.9407 on the DRIONS, DRISHTI, and RIMONE-v3 datasets, respectively. The results obtained at the end of the study show that the proposed model is robust on these datasets and is competitive with other studies in the literature. In conclusion, we showed that the proposed model can be used effectively in glaucoma diagnosis.

Hand fractures, particularly phalangeal fractures, are often subtle, concealed, and complex, posing significant challenges for accurate and efficient diagnosis with traditional radiography. To address the limitations of existing deep learning methods in handling complex anatomy, multi-view uncertainty, and causal discrimination, we propose a novel two-stage context-aware multi-view fusion framework for precise fracture classification and localization. In the first stage, a Context-Aware Multi-View Classification Network (CAMVC-Net) is developed. A Dual-path Semantic-Spatial Alignment (DSSA) module aligns low-level spatial details with high-level semantics, enhancing fine-grained fracture discrimination. A Context Aware Pyramid (CAP) module captures multi-scale context, while a Counterfactual Attention Learning (CAL) loss guides the network to focus on discriminative regions. Multi-view uncertainty is modeled using a Dirichlet distribution, and decision-level fusion is achieved by Dempster's rule to mimic clinical reasoning. In the second stage, a Detail-aware Fracture Localization Network (DFL-Net) is designed. To adapt to irregular fracture geometry, DFL-Net integrates deformable convolutions and incorporates DSSA into the feature pyramid to preserve fine spatial details during deep downsampling. Experiments on the MURA and clinical datasets demonstrate strong performance: the classification model achieved an accuracy of 92.89%, precision of 93.05%, recall of 96.96%, and F1 score of 94.97%. The localization network obtained 73.8% AP₅₀ and 43.8% AP₇₅, with a 10.4% improvement in AP₇₅ over Faster R-CNN. These results indicate that the proposed framework provides an accurate and efficient tool for computer-aided fracture diagnosis, with the potential to reduce misdiagnosis and improve clinical decision-making efficiency.

In CT imaging, the size, shape, margin, and density of pulmonary nodules are used to determine whether they are benign or malignant. However, pulmonary nodules often exhibit challenging characteristics in CT images, such as irregular shapes, and tiny nodules are prone to being overlooked. In addition, the density of nodules may be similar to that of surrounding tissues. When the nodules are small or close to the pulmonary wall, their resolution is low, making them difficult to distinguish. Because of these characteristics, it is quite challenging to segment pulmonary nodules automatically. This research proposes CCEMSS-Unet++, a medical image segmentation network that enhances feature fusion between local details and global context through a multi-scale design. It includes the CCEMS module and the SE attention mechanism. The SE module mitigates false negatives, particularly for small nodules; the CCEMS module is used to extract local information and connect global context, improving segmentation accuracy. This study used two datasets: the publicly available LIDC-IDRI dataset and a dataset collected from the CT Department of Yan'an University Affiliated Hospital. Compared with Unet++, the proposed model improves IoU and Dice by 2.62% and 2.15% on the LIDC-IDRI dataset, and by 5.10% and 7.11% on our dataset, respectively. In terms of segmentation accuracy and generalization, the experimental findings demonstrate that this approach outperforms other networks, including handling nodules with low resolution. The efficacy of the proposed enhancement method is further supported by the ablation experiments.

Precise segmentation of critical retinal structures, including the optic cup, optic disc, and vascular bifurcation sites, is essential for the early identification and management of severe ocular conditions such as glaucoma, diabetic retinopathy, hypertensive retinopathy, and age-related macular degeneration. This study proposes a hybrid multi-attention residual-based UNet (MRUNet+), a deep learning (DL) model developed to address the complex nature of retinal image processing using an advanced attention mechanism to enhance segmentation precision. MRUNet+ was trained on many high-quality datasets, including DRIVE, CHASE_DB1, Drishti-GS1, RIM-ONE and REFUGE1, enabling it to address prevalent imaging issues, such as variations in contrast, resolution and the presence of abnormalities. The model has exhibited markedly enhanced performance compared to recent methodologies, with improved accuracy metrics, Dice coefficients and intersection over union (IoU), particularly in low-contrast or complex anatomical regions. Due to its adaptability and precision, MRUNet+ has the potential to function as a pivotal instrument in clinical settings, providing automated retinal assessments that aid in the prompt detection of various retinal and vascular pathologies, thereby improving patient management.

Accurate grading of invasive ductal carcinoma (IDC) remains hard even for the trained pathologist. The patient's therapeutic schedule depends on the observed grade, emphasizing the importance of accurate tissue grading. We propose a novel architecture called Twin Tokens Talking-heads Class Attention Net (T2TCAN), designed to grade IDC images. Based on the transformer architecture, this design enhances the model's ability to capture local and global tissue features. This model was trained on the PathoIDCG dataset, outperforming the existing methods in terms of standard metrics. Further, introducing class and twin tokens later in the model, along with the talking-heads class attention and per-channel scaling, made the T2TCAN model efficient at distinguishing the tissue grades. Our model outperformed conventional CNN-based IDC grading approaches. We report the highest 98.82% accuracy, AUC (0.99), precision (98.83%), recall (98.80%), and F1-score (98.81%) on the PathoIDCG dataset. We plotted class attention maps for the TCT layers to provide insights into the model's decision-making process. These attention maps are also helpful for pathologists to understand the discriminative features that significantly contribute to prognostic scoring. Further, our model also performed well on a separate IDC grading dataset, producing state-of-the-art results and thus proving its efficacy. The designed model claims its adaptability and suitability for integration into the healthcare workflow for IDC grading.