International Journal of Imaging Systems and Technology最新文献

{"title":"A Classification Network With Coordinate and Class-Specific Residual Attention for Accurate Detection of Heart Failure-Related Findings in Chest X-Rays","authors":"Yi Li,&nbsp;Dengao Li,&nbsp;Jumin Zhao,&nbsp;Liangquan Yan,&nbsp;Xiaole Fan","doi":"10.1002/ima.70408","DOIUrl":"https://doi.org/10.1002/ima.70408","url":null,"abstract":"<div>\u0000 \u0000 <p>Heart failure (HF) is a life-threatening condition that poses a major global health burden. Accurate and automated detection of HF-related manifestations—namely cardiomegaly, effusion, and edema—in chest X-rays (CXRs) is of considerable clinical value for timely treatment. However, CXR abnormalities vary markedly in spatial scale, and the strong interdependencies among disease categories further complicate recognition, often resulting in degraded classification performance. To address these challenges, we introduce a compact, fully supervised classification network termed CFC, which integrates three complementary modules: (1) Coordinate Attention (CA) embedded in the backbone to enhance long-range dependency modeling while preserving precise positional information; (2) a top-down multiscale Feature-wise Layer Fusion (FLF) pathway to effectively integrate semantic and spatial cues; and (3) a Class-Specific Residual Attention (CSRA) head that concentrates on the most discriminative regions for each pathology. The proposed architecture was extensively evaluated on the CheXpert dataset, the ChestX-ray14 dataset, and a proprietary in-house clinical dataset. Compared with state-of-the-art methods, our model achieved mean AUC scores of 0.9085, 0.9068, and 0.9253, respectively. These findings demonstrate that the CFC framework offers an effective and computationally efficient solution for HF diagnosis across both public benchmarks and real-world clinical environments.</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 5","pages":""},"PeriodicalIF":3.0,"publicationDate":"2026-07-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148616201","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"ImRecUnet: Colon Polyp Segmentation Based on Adaptive Improved Recalling U-Net Context","authors":"Nguyen Thanh Binh,&nbsp;Vo Thi Hong Tuyet","doi":"10.1002/ima.70391","DOIUrl":"https://doi.org/10.1002/ima.70391","url":null,"abstract":"<div>\u0000 \u0000 <p>Colon polyp segmentation is important in the early detection of colorectal cancer. Advances in artificial intelligence and computer-aided diagnosis systems have dramatically improved the capabilities of colonoscopy by enhancing the segmentation and detection of colonic polyps. Accurate segmentation is essential, as it allows for the differentiation of polyps from surrounding tissues and aids in assessing their potential malignancy. This paper proposed a method for segmenting colon polyp images by improving the Recalling U-Net model based on the original U-Net and U-Net++ models. The new model aims to take full advantage of multiscale features by introducing full-scale skip connections, combining low-level details with high-level semantics from full-scale feature maps, but with fewer parameters. The model can develop deep supervision to learn hierarchical representations from full-scale aggregated feature maps in order to optimize the combined loss function to enhance organ boundaries. Specifically, the proposed method improves four major changes compared to Recalling U-Net: separate down sampling process, improved encoding path, upgraded bridge and improved recalling block layers. These changes help improve the learning ability and efficiency of the proposed method, resulting in more accurate segmentation results. The proposed method is tested on the Kvasir-SEG, CVC-ClinicDB and CVC-ColonDB datasets with evaluation metrics such as Dice coefficient, Intersection over Union and accuracy. Experimental results presented that the proposed method achieved an accuracy of 98.992%, 97.628%, and 97.695% on these datasets, respectively, and outperforms the results of several recent methods on the same dataset.</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":3.0,"publicationDate":"2026-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148282638","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Histopathology-Driven Multimodal Modeling for Lymph Node Metastasis Prediction and Survival Stratification in Lung Squamous Cell Carcinoma","authors":"Rong Fu,&nbsp;Lili Wang,&nbsp;Ning Wang,&nbsp;Qi Yan,&nbsp;Xingjiang Li,&nbsp;Tianshu Wang,&nbsp;Jianzhou Liu,&nbsp;Mingxiu Kong,&nbsp;Kaiqi Zhang,&nbsp;Xiang Li,&nbsp;Hongqun Tang","doi":"10.1002/ima.70359","DOIUrl":"https://doi.org/10.1002/ima.70359","url":null,"abstract":"<div>\u0000 \u0000 <p>To develop a multimodal computational pathology framework integrating handcrafted morphology and deep Vision Transformer (ViT) features to improve lymph node metastasis prediction and survival risk stratification in early-stage lung squamous cell carcinoma. Whole-slide images (WSIs) from 211 TCGA LUSC patients and an external cohort of 150 cases underwent standardized stain normalization, adaptive multiresolution patch extraction, and stringent artifact removal. Handcrafted morphological features were extracted using a customized CellProfiler pipeline, and deep embeddings were obtained from a fine-tuned ViT model. After ICC reliability filtering and correlation reduction, LASSO, mutual information (MI), and Boruta selection strategies were applied. A unified early-fusion vector was constructed and evaluated using SVM, XGBoost, TabNet, and TabTransformer classifiers. Survival analyses employed Cox-LASSO, Elastic Net, Random Survival Forests, and Gradient Boosting Survival models, with performance assessed using AUC, C-index, IBS, and time-dependent AUC. Deep ViT-derived features outperformed handcrafted morphology across all metrics (external AUC 0.82 vs. 0.71), and fusion significantly enhanced predictive performance, achieving test and external AUCs up to 0.94 and 0.92, respectively. The LASSO-TabTransformer fusion pipeline provided the strongest metastasis discrimination, with recall approaching 87%. Survival modeling demonstrated similarly substantial improvements: fusion-based Gradient Boosting Survival achieved C-indices of 0.91 (test) and 0.88 (external), with 1-year AUCs of 0.92 and 0.89. FDR-validated imaging covariates remained independently prognostic, and Kaplan–Meier curves confirmed robust stratification of LNM+ versus LNM− groups. Multimodal fusion of deep and handcrafted features produces highly accurate, generalizable models for LNM prediction and survival stratification in LUSC, offering a powerful tool for preoperative decision support.</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":3.0,"publicationDate":"2026-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148282301","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Portable AI-Enabled Intraoral Imaging System for Real-Time Detection and Staging of Dental Caries","authors":"Hafiz Muhammad Salman Ajmal,&nbsp;Muhammad Faisal Mehmood,&nbsp;Muhammad Umair Ahmad Khan,&nbsp;Zabdur Rehman,&nbsp;Ezzat Fatima,&nbsp;Muhammad Imran Iqbal,&nbsp;Waqar Muhammad,&nbsp;Izaz Raouf","doi":"10.1002/ima.70395","DOIUrl":"https://doi.org/10.1002/ima.70395","url":null,"abstract":"<div>\u0000 \u0000 <p>Dental caries is one of the most common oral diseases and it requires advanced imaging techniques for early and accurate detection. This paper presents a portable AI-enabled intraoral imaging system inspired by the YOLOv8 deep learning model and equipped with a low-cost hardware Raspberry Pi 4 and intraoral endoscopy camera. Intraoral photos that were clinically obtained were annotated as per the American Dental Association ADA criteria of caries and utilized in the training and validation of the system. The optimized model achieved 81.5% accuracy, with precision, recall, and F1-scores of 82%, 81%, and 81%, respectively, for stage-specific detection of caries. Real-time performance and low computational overhead were demonstrated, highlighting the feasibility of deploying embedded AI for point-of-care imaging. This hardware–software co-designed system introduces a cost-effective and scalable diagnostic platform for intraoral applications, and its framework is extendable to other medical imaging tasks requiring portable, AI-driven solutions.</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":3.0,"publicationDate":"2026-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148282582","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"We Must Be Honest About the Clinical Value of AI: When High Performance Metrics Might Hide Inconvenient Biases","authors":"Mohamed L. Seghier","doi":"10.1002/ima.70393","DOIUrl":"https://doi.org/10.1002/ima.70393","url":null,"abstract":"&lt;p&gt;This editorial highlights a concerning trend where the discussion about the clinical ‘value’ of AI-based solutions for healthcare is being dominated by performance metrics (i.e., number-centric decisions) that do not tell the whole story and can skew evidence. As researchers, we strive to make quantitative metrics as objective and bias-free as possible, but the history of science has also shown us that misuse or abuse of these metrics can manipulate evidence or distort validity. In this context, it could be informative to make a parallel between how researchers are using model performance metrics today with how statistics were (mis)used in the past to change clinical practices that then turned out to be based on inflated numbers and biased evidence.&lt;/p&gt;&lt;p&gt;Model performance metrics are becoming pervasive in the field of digital health and medical AI, more often defined as standards for good and useful models [&lt;span&gt;1-3&lt;/span&gt;]. It is well-documented that not all metrics are equal, and authors are always asked to select the right metric given their model's scope and data type [&lt;span&gt;1, 4, 5&lt;/span&gt;]. What we are concerned about here is how such metrics are sometimes used to inflate AI models' capacity and efficiency, thereby promoting them as ready for real-world clinical applications. Indeed, operating under pressure to adopt AI, clinicians and policy makers are sometimes being sold solutions based on high performance metrics without full appreciation of other factors and dimensions not always well captured by those reductionist metrics. Furthermore, the current research ecosystem, skewed toward positive effects [&lt;span&gt;6, 7&lt;/span&gt;], typically rewards overconfident claims and remarkable metrics, and the history of statistics reminds us that numbers can, intentionally or inadvertently, be used to mislead or to obscure. Medical AI is now at the same ethical crossroads: beyond the performance metrics, one should ask what exactly AI tools are solving or adding to clinical workflows [&lt;span&gt;8&lt;/span&gt;] and how to enforce clear adherence to principles of transparency, fairness, and accountability [&lt;span&gt;9&lt;/span&gt;].&lt;/p&gt;&lt;p&gt;AI is transforming healthcare on an unprecedented scale [&lt;span&gt;10&lt;/span&gt;]. Recent evidence points to the fact that the regulatory and commercial cycle for medical AI is becoming shorter than the cycle traditionally expected for other medical technologies. On average, it takes 17 years to translate evidence into products and practice in the healthcare sector [&lt;span&gt;11, 12&lt;/span&gt;]. However, already AI-based solutions are enjoying a much faster approval track, thereby allowing AI to disrupt clinical practice at a quicker pace than previous disruptive technologies. For instance, there are already almost 100 AI tools for radiology available in the European market, approved by the FDA (under Class II) and adhering to the European Medical Device Regulations [&lt;span&gt;13&lt;/span&gt;]. Consequently, AI tools are pushed toward implementation on timelines th","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/ima.70393","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148238113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing Breast Cancer Diagnosis: Harnessing Deep Learning With Pretrained and Hybrid CNNs for Efficient Mammogram Analysis","authors":"Mayank Arora,&nbsp;Poonam Saini,&nbsp;Kanu Goel","doi":"10.1002/ima.70378","DOIUrl":"https://doi.org/10.1002/ima.70378","url":null,"abstract":"<div>\u0000 \u0000 <p>Deep learning has played a significant role in improving breast cancer detection using mammographic images. However, several limitations persist in existing studies, including imbalance in class distribution, inconsistent use of datasets, and dependence on individual convolutional neural network (CNN) models that may not generalize well across different data settings. To address these challenges, this study proposes a structured framework for breast cancer classification using pretrained CNN architectures and hybrid learning models. All experiments are conducted using the DDSM dataset to ensure consistency and avoid data-related biases. A standardized preprocessing pipeline is applied to all images, incorporating intensity normalization, contrast enhancement, and noise reduction techniques to improve image quality. To further enhance model performance, data augmentation is applied selectively to the training set to mitigate class imbalance. Feature representations extracted from CNN models such as ResNet, DenseNet, VGG, and Inception are combined with machine learning classifiers including Random Forest and XGBoost. The experimental results indicate that hybrid approaches achieve more consistent and reliable performance compared to conventional CNN-based classification. This study demonstrates that integrating deep feature extraction with ensemble learning methods can improve the robustness of mammogram classification while maintaining methodological rigor. The performance gains are influenced by the ROI-based dataset characteristics and controlled experimental setup, and further validation on diverse datasets is required. The results are influenced by ROI-based inputs and controlled experimental conditions.</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148238107","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"3D-HAFNet: A Novel Hierarchical Aggregation Feature Network With Triple Attention for Early Alzheimer’s Disease Detection From Structural MRI","authors":"Zi-Hao Wang,&nbsp;Zhen-Dong Niu,&nbsp;Wei Yi,&nbsp;Rui Ren,&nbsp;Xiang-Qin Ji","doi":"10.1002/ima.70388","DOIUrl":"https://doi.org/10.1002/ima.70388","url":null,"abstract":"<div>\u0000 \u0000 <p>Early detection of Alzheimer's disease (AD) during the mild cognitive impairment (MCI) stage remains clinically challenging despite its critical importance for therapeutic intervention. We present 3D-HAFNet, a deep learning framework for early AD detection using structural magnetic resonance imaging (sMRI) that addresses two fundamental challenges: capturing multi-scale pathological patterns and exploiting directional information inherent in anisotropic brain data. The model incorporates two key innovations: (1) a Hierarchical Aggregation Feature Pyramid (HAFP) enabling bidirectional multi-scale feature fusion across network layers, and (2) a Triple Attention (TA) mechanism extending traditional channel and spatial attention with directional attention tailored for 3D neuroimaging. Experiments on 1247 subjects from the ADNI dataset demonstrate superior performance with 92.3% overall accuracy for three-class classification (NC/MCI/AD). Binary classification tasks achieve strong performance: AD vs. NC (94.7% accuracy, AUC: 0.967), MCI vs. NC (85.7% accuracy, AUC: 0.918), and AD vs. MCI (88.3% accuracy, AUC: 0.942), outperforming state-of-the-art methods. Furthermore, external validation on an independent OASIS-3 cohort confirms the model's robust generalization capability. Ablation studies validate that adding HAFP to the baseline improves accuracy by 20.7 percentage points, and further incorporating TA yields an additional 4.5 percentage-point gain. The directional attention mechanism represents a novel approach for exploiting anatomical plane-specific information in 3D neuroimaging analysis. The current implementation remains limited to single-time-point sMRI and requires longitudinal, multimodal, prospective, and lightweight-deployment validation before routine clinical use.</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148237697","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Factors Affecting Completion Rate and Bowel Cleansing Quality in a Colon Capsule Endoscopy Cohort: Comparing AI and Manual Evaluations","authors":"Benedicte Schelde-Olesen,&nbsp;Gunnar Baatrup,&nbsp;Jan-Matthias Braun,&nbsp;Jürgen Herp,&nbsp;Lasse Kaalby,&nbsp;Anastasios Koulaouzidis,&nbsp;Thomas Bjørsum-Meyer,&nbsp;Ulrik Deding,&nbsp;the CareForColon2015 group and the AICE consortium","doi":"10.1002/ima.70386","DOIUrl":"https://doi.org/10.1002/ima.70386","url":null,"abstract":"<p>In the trial CareForColon2015 (CFC2015) we included more than 2000 colon cancer screening participants for colon capsule endoscopy (CCE). We detected distinct fluctuations in the rate of complete investigations defined as CCE with both complete transit and adequate bowel cleansing. We aimed to investigate possible contributing factors to incomplete investigations and compare fluctuations in manual and artificial intelligence (AI) generated cleansing evaluations. We retrieved CCE videos and relevant information from CFC2015. We considered the following factors as possible contributors to completion rate fluctuations: different outpatient clinics for CCE, individual nurses instructing participants, number of participants at the information session, timing of the last dose of bowel preparation, timing of capsule ingestion, time gap between the last dose of bowel preparation and capsule ingestion, manual CCE reader-cleansing evaluations. During the trial, the prokinetic prucalopride was added to the regimen. We compared the completion rates between different time periods of the trial. Monthly completion rates ranges from 56.6% to 75.7%. The fluctuations were caused by variation in cleansing quality and not in the proportion of complete transit. Early capsule ingestion showed increased odds of a complete investigation. The timing of the last dose of bowel preparation was associated with cleansing quality. Fluctuations in cleansing quality were detected by manual reading in all colonic segments but by the AI algorithm predominantly in the right colon. Timing of bowel preparation and capsule ingestion is an important factor for the bowel cleansing quality in CCE. Using AI for cleansing quality monitoring in a CCE cohort can be an important measure for early identification of changes.</p><p><b>Trial Registration:</b> ClinicalTrials.gov: NCT04049357</p>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/ima.70386","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148237757","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Predicting Mild Cognitive Impairment to Alzheimer's Disease Progression by Explainable Multi-Modal Framework Using Deep and Machine Learning Hybrid Model","authors":"Soheil Zarei,&nbsp;Mohsen Saffar,&nbsp;Peyman Hassani Abharian,&nbsp;Reza Shalbaf,&nbsp;Ahmad Shalbaf","doi":"10.1002/ima.70390","DOIUrl":"https://doi.org/10.1002/ima.70390","url":null,"abstract":"<div>\u0000 \u0000 <p>Accurate prediction of the progression from mild cognitive impairment (MCI) to Alzheimer's disease (AD) is essential for early intervention and effective disease management. This study introduces an interpretable multi-modal framework that integrates structural Magnetic Resonance Imaging (sMRI) features and neurocognitive clinical assessments through a hybrid deep learning and machine learning approach to predict progressive MCI (pMCI) conversion up to 3 years in advance. Cross-sectional data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) were employed. Spatial features were extracted from sMRI scans using a three-dimensional convolutional neural network (3D CNN), while statistically selected neurocognitive assessments constituted the clinical feature set. These features were fused and classified using machine learning models, including Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), and Gradient Boosting Machine (GBM). Model interpretability was evaluated using SHapley Additive Explanations (SHAP). The proposed multi-modal framework achieved a classification accuracy of 88.11%, an area under the curve (AUC) of 0.889, a precision of 82.33%, a sensitivity of 90.52%, and a specificity of 78.59%. The GBM classifier demonstrated superior performance by effectively integrating imaging and clinical features. SHAP analysis identified critical biomarkers influencing prediction outcomes, and the model exhibited strong generalizability across diverse clinical sites and imaging protocols. Consequently, this study presents a robust and interpretable framework for the early prediction of MCI conversion to AD. By combining deep learning-derived imaging features with neurocognitive assessments, the proposed method offers a promising tool to support early diagnosis and targeted intervention strategies in clinical practice.</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-06-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148237742","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Segmentation of Multiple Sclerosis Lesions in MRI Using U-Net With a Wavelet Transformer-Based Attention Mechanism","authors":"Ali Alijamaat,&nbsp;Seyed Mohsen Mirhosseini","doi":"10.1002/ima.70374","DOIUrl":"https://doi.org/10.1002/ima.70374","url":null,"abstract":"<div>\u0000 \u0000 <p>Multiple sclerosis (MS) is a chronic autoimmune central nervous system disorder characterized by the immune system's mistaken attack on myelin, leading to disrupted neural signal transmission and diverse neurological impairments. Early and accurate diagnosis is crucial for effective disease management. While lesion detection in MRI scans plays a fundamental role in diagnosis, the small size and scattered distribution of lesions present significant challenges. Deep learning-based automatic segmentation has shown promise in improving diagnostic accuracy. U-Net is a widely adopted architecture for medical image segmentation, yet its dense decoder layers hinder high-resolution feature extraction. This study introduces an enhanced U-Net architecture incorporating a local attention mechanism and wavelet transform within skip connections, complemented by a spatial attention mechanism in the bottleneck. These modifications enable the model to capture both local and global image features, leading to more precise lesion segmentation. The proposed method was evaluated on the ISIB2015 dataset, demonstrating superior performance compared to conventional U-Net and similar approaches in terms of dice similarity coefficient (DSC) and positive predictive value (PPV).</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"36 4","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-06-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148237868","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}