International Journal of Imaging Systems and Technology最新文献

To develop and validate a multimodal machine learning model for opportunistic osteoporosis screening in postmenopausal women using dental periapical radiographs. This retrospective multicenter study analyzed 3885 periapical radiographs paired with DEXA-derived T-scores from postmenopausal women. Clinical, handcrafted radiomic, and deep features were extracted, resulting in a fused feature set. Radiomic features (n = 215) followed Image Biomarker Standardization Initiative (IBSI) guidelines, and deep features (n = 128) were derived from a novel attention-based autoencoder. Feature harmonization used ComBat adjustment; reliability was ensured by intra-class correlation coefficient (ICC) filtering (ICC ≥ 0.80). Dimensionality was reduced via Pearson correlation and LASSO regression. Four classifiers—logistic regression, random forest, multilayer perceptron, and XGBoost—were trained and evaluated across stratified training, internal, and external test sets. A logistic regression model was selected for clinical translation and nomogram development. Decision curve analysis assessed clinical utility. XGBoost achieved the highest classification performance using the fused feature set, with an internal AUC of 94.6% and external AUC of 93.7%. Logistic regression maintained strong performance (external AUC = 91.3%) and facilitated nomogram construction. Deep and radiomic features independently outperformed clinical-only models, confirming their predictive strength. SHAP analysis identified DEXA T-score, age, vitamin D, and selected radiomic/deep features as key contributors. Calibration curves and Hosmer–Lemeshow test (p = 0.492) confirmed model reliability. Decision curve analysis showed meaningful net clinical benefit across decision thresholds. Dental periapical radiographs can be leveraged for accurate, non-invasive osteoporosis screening in postmenopausal women. The proposed model demonstrates high accuracy, generalizability, and interpretability, offering a scalable solution for integration into dental practice.

Lung cancer remains the leading cause of cancer-related incidence and mortality worldwide. Early detection of lung nodules is crucial for significantly reducing the risk of lung cancer. However, due to the high similarity in CT image features between lung nodules and surrounding normal tissues, nodules are often missed or misidentified during the detection process. Moreover, the diverse types and morphologies of nodules further complicate the development of a unified detection approach. To address these challenges, this study proposes a novel Feature Reconstruction-guided Multi-Scale Attention Network (FRMANet). Specifically, a Refined Feature Reconstruction Module is designed to effectively suppress redundant information while preserving essential feature representations of nodules, ensuring high sensitivity and enhanced representation capability for nodule regions during feature extraction. Additionally, a Multi-scale Feature Enhancement Attention mechanism is introduced, which utilizes an attention-based fusion strategy across multiple scales to fully capture discriminative features of nodules with varying sizes and shapes. Experimental results on the LUNA16 dataset demonstrate that the proposed FRMANet achieves superior detection performance, with a mAP of 0.894 and an F1 score of 0.923, outperforming existing state-of-the-art methods.

Identifying the primary tumor origin is a critical factor in determining treatment strategies for brain metastases, which remain a major challenge in clinical practice. Traditional diagnostic methods rely on invasive procedures, which may be limited by sampling errors. In this study, a dataset of 200 patients with brain metastases originating from six different cancer types (breast, gastrointestinal, small cell lung, melanoma, non-small cell lung, and renal cell carcinoma) was included. Radiomic features were extracted from different magnetic resonance images (MRI) and selected using the Kruskal–Wallis test, correlation analysis, and ElasticNet regression. Machine learning models, including support vector machine, logistic regression, and random forest, were trained and evaluated using cross-validation and unseen test sets to predict the primary origins of metastatic brain tumors. Our results demonstrate that radiomic features can significantly enhance classification accuracy, with AUC values reaching 0.98 in distinguishing between specific cancer types. Additionally, survival analysis revealed significant differences in survival probabilities across primary tumor types. This study utilizes a larger, single-center cohort and a standardized MRI protocol, applying rigorous feature selection and multiple machine learning classifiers to enhance the robustness and clinical relevance of radiomic predictions. Our findings support the potential of radiomics as a non-invasive tool for metastatic tumor prediction and prognostic assessment, paving the way for improved personalized treatment strategies. Radiomic features extracted from MRI images can significantly enhance the prediction of the main origin of the metastatic tumor types in the brain, thereby informing treatment decisions and prognostic assessments.

Patient demographic prediction involves estimating age, gender, ethnicity, and other personal characteristics using X-rays. This can help in personalized medicine and improved healthcare outcomes. It can assist in automated diagnosis for some diseases that exhibit age and gender-specific prevalence. It can also help in forensic science to identify individuals when demographic information is missing. Insights from deep learning can verify the gender and age of self-reported individuals through chest X-rays (CXRs). In this proposed work, we have deployed an artificial intelligence (AI) enabled model which focuses on two tasks: gender classification and age prediction from CXRs. For gender classification, the model combines ResNet-50 (CNN) and Vision Transformer (ViT) to leverage both local feature extraction and global contextual understanding for predicting gender and is called ViTCXRResNet. The model was trained and validated on an Amazon Web Services (SPR) dataset of 10702 images, split with an 80–20 ratio, which was evaluated with classification metrics to determine the model's behavior. For age prediction, extracted features from ResNet-50 were used with dimensionality reduction through principal component analysis (PCA). A fully connected feedforward neural network was trained on the reduced feature set to predict age. The classification and regression model achieves accuracy results of 93.46% for gender classification and 0.86 for the R² score for age prediction on the SPR dataset. For visual interpretation, explainable AI (Gradient-weighted Class Activation Mapping) was utilized to visualize and find out which parts of the image are prioritized for classifying gender. The proposed model yields high classification accuracy in gender detection and significant accuracy in age prediction. The model shows competitive accuracy compared to existing methods. Further, the demographic prediction stability of the model was proven on two different ethnic groups, such as the Japanese Society of Radiological Technology (JSRT) and Montgomery (USA) datasets.

As the first line of defense in the human immune system, the skin is highly susceptible to environmental toxins. Melanoma, the most lethal type of skin cancer, is characterized by high mortality and a strong tendency for metastasis. It can sometimes originate from pre-existing nevi, particularly dysplastic nevi. Early identification is crucial for improving patient survival rates. However, traditional skin lesion detection faces challenges due to image quality limitations, dataset imperfections, and the complexity of lesion features. This study proposes the DenseNet201SA++ model, which uses image augmentation techniques and the soft attention mechanism to optimize dermoscopy image quality and automatically capture critical features. Experiments on the HAM10000 dataset with 10,015 dermoscopic images, focusing on binary classification (melanoma vs. nevus), show that the DenseNet201SA++ model achieves significant performance gains, with improvements in precision, recall, F1-score, and accuracy of at least 7.2%, 14.7%, 12.7%, and 14.7% compared to baseline networks. The proposed soft attention-guided feature fusion in DenseNet201SA++ addresses feature redundancy in traditional attention mechanisms, achieving superior performance in distinguishing Mel from Nv, while the DenseNet201 backbone shows distinct advantages. Ablation studies confirm the significant role of data augmentation. The integrated DenseNet201SA++ model achieves robust results with precision, recall, F1-score, and accuracy all reaching 0.983, complemented by an AUC of 0.993. These metrics demonstrate the model's exceptional balance between discriminative power and generalization capability, validating the effectiveness of our proposed architecture.

Accurate classification of viral DNA sequences is essential for tracking mutations, understanding viral evolution, and enabling timely public health responses. Traditional alignment-based methods are often computationally intensive and less effective for highly mutating viruses. This article presents a machine learning framework for classifying DNA sequences of COVID-19 variants using K-mer-based tokenization and vectorization techniques inspired by Natural Language Processing (NLP). DNA sequences corresponding to Alpha, Beta, Gamma, and Omicron variants are obtained from the Global Initiative on Sharing All Influenza Data (GISAID) database and encoded into feature vectors. Multiple classifiers, including Extra Trees, Random Forest, Support Vector Classifier (SVC), Decision Tree, Logistic Regression, Naive Bayes, K-Nearest Neighbor (KNN), Ridge Classifier, Stochastic Gradient Descent (SGD), and XGBoost, are evaluated based on accuracy, precision, recall, and F1-score. The Extra Trees model achieved the highest accuracy of 93.10% $��\pm �$ 0.42, followed by Random Forest with 92.60% $��\pm �$ 0.38, both demonstrating robust and balanced performance. Statistical significance tests confirmed the robustness of the results. The results validate the effectiveness of K-mer-based encoding combined with traditional machine learning models in classifying COVID-19 variants, offering a scalable and efficient solution for genomic surveillance.

The multi-modal medical image fusion (M3IF) combines the required and important information from different medical imaging modalities (computed tomography [CT], magnetic resonance imaging (MRI), positron emission tomography [PET], and single photon emission computed tomography [SPECT]) to provide single informative image. M3IF provides enhanced patient diagnosis, and precise treatment planning. This paper proposes a hybrid M3IF where input medical images are decomposed using stationary wavelet transform (SWT) into low-frequency components (LFCs) and high-frequency components (HFCs). The LFCs and HFCs are fused using energy- and contrast-based metrics. And later reconstruction is performed using inverse SWT (ISWT). The total variation minimization (TVM) using Chambolle's algorithm is applied as a post-refinement operation to reduce noise and preserves the fine details. In this paper, the proposed methodology is termed as M3IF-(SWT-TVC), Here, the acronym TVC is the combination of TVM using Chambolle's algorithm. TVM refinement process is an iterative approach, with the fusion outcomes of M3IF-(SWT-TVC) assessed over a predefined 100 iterations. The TVM, and SWT are blended to balance smoothness and structural details. The final fusion results obtained through M3IF-(SWT-TVC) are evaluated against several prominent non-traditional methods. Based on both visual quality and quantitative metric analysis, it is observed that M3IF-(SWT-TVC) outperforms all the methods used for comparison.

This study explores the effectiveness of deep learning methodologies in the detection and classification of lumbar disc intensity using MRI scans. Initially, region-based deep learning frameworks, including Faster R-CNN and Mask R-CNN with different backbones such as ResNet50 and ResNet101 are evaluated. Results demonstrated that backbone selection significantly impacts model performance, with Mask R-CNN combined with ResNet101 achieving a remarkable [email protected] (AP50) of 99.83%. In addition to object detection models, Transformer-based classification architectures, including MaxViT, Vision Transformer (ViT), a Hybrid CNN-ViT model, and Fine-Tuned Enhanced Pyramid Network (FT-EPN), are implemented. Among these, the Hybrid model achieved the highest classification accuracy (83.1%), while MaxViT yielded the highest precision (0.804). Comparative analyses highlighted that while Mask R-CNN models excelled in segmentation and detection tasks, Transformer-based models provided effective solutions for direct severity classification of lumbar discs. These findings emphasize the critical role of both backbone architecture and model type in optimizing diagnostic performance. The study demonstrates the potential of integrating region-based and Transformer-based models in advancing automated lumbar spine assessment, paving the way for more accurate and reliable medical diagnostic systems.

Deep learning-based methods have achieved great progress in CT image segmentation in recent years. However, the lack of unified large-scale datasets, unbalanced categories of segmented images, blurred boundaries between infected and healthy regions, and different sizes and shapes of lesions have led to the fact that the existing methods still have challenges in further improving the segmentation accuracy in medical applications. In this work, a novel network with spectrum Transformer and triplet attention for CT image segmentation is proposed. The computed spectrum Transformer module and the triplet attention module (TAM) are fused in a parallel way utilizing the parallel hybrid fusion module (PHFM), which extracts global and local contextual information from sequence features and cross-dimensional features. A spectrum Transformer block (STB) is proposed, which utilizes the fast Fourier transform (FFT) to learn the weights of each frequency component in the spectral space. Extensive comparison experiments are conducted on both the COVID-19-Seg dataset and Mosmeddata, showing that the proposed network has better accuracy than most existing methods for CT image segmentation tasks. In particular, the proposed model achieves improvement by 1.29% and 2.28% in terms of DSC, respectively, by 1.15% and 1.35% in terms of mIoU. SEN metrics increase by 1.45% and 1.48%, respectively. PRE also achieves the best results, showing its significant advantage in accurately segmenting medical images. SPE also shows quite good results in both datasets. Ablation studies show that the proposed STB and TAM modules improve the segmentation performance significantly.