International Journal of Imaging Systems and Technology最新文献

The employment of artificial intelligence methods in computer-assisted diagnosis systems is critical for colorectal cancer survival analysis and prognosis. However, due to the low prediction accuracy of single-modal data research and the complexity of multimodal data fusion methods, the current study's effect on colorectal cancer is minimal. To address this issue, the authors offer a multimodal cross attention fusion (MMCAF) technique for predicting colorectal cancer survival status. First, feature engineering is used to create feature sets for every mode and to address the heterogeneity of multimodal data. Second, a three-mode fusion technique is used to allocate weight to single-mode and multimodal features via channels and cross-attention processes. Lastly, the cross-entropy loss function is minimized in order to estimate the classification survival. The experimental results reveal that the MMCAF approach predicts survival states with 97.73% accuracy and an area under the receiver operating characteristic curve (AUC) of 0.99. When compared to the best outcome of other fusion algorithms (feature concatenation), the prediction accuracy increases by about 6 percentage points, while the AUC increases by 7 percentage points. This finding thoroughly demonstrates MMCAF's efficacy in predicting colorectal cancer survival.

Skin disorders are common and require diagnosis and treatment in a timely manner. In traditional diagnostics, great demands are made on the time and interpretation of the results. To cope with this, we introduce YOLOv11, an enhanced deep learning model designed for skin disease detection and classification. The model integrates EfficientNetB0 as the backbone for feature extraction and ResNet50 in the head for robust classification and localization. Our model was trained on a dataset of 10 common skin diseases to ensure robustness and accuracy; we were able to classify the diseases with a mean Average Precision (mAP) of 89.8%, a precision of 90%, and a recall of 88% on the test dataset. This model was developed in the form of a web application based on Streamlit, which was used for easy uploading of pictures by both clinicians and patients for threshold diagnostics. This upsurge in technology allows for treatment without visitation, making skin disease diagnosis more dynamic.

Efficient and accurate segmentation of brain tumors in multimodal magnetic resonance imaging (MRI) is crucial for clinical diagnosis and treatment planning. Traditional methods tend to concentrate solely on feature extraction from individual modalities, overlooking the substantial potential of multimodal feature fusion in enhancing segmentation performance. In this paper, we present a novel method that not only integrates salient features from different modalities strategically but also takes into account the constraints imposed by limited computational resources, ensuring both accuracy and efficiency. Two key modules, the attention-guided cross-modality fusion module (ACFM) and the hierarchical asymmetric convolution module (HACM), were designed to leverage the distinct modalities and the varying information focuses found within different dimensions. The ACFM is based on a transformer framework, utilizing self-attention and cross-attention mechanisms. These mechanisms enable the capture of both local and global dependencies within and between different MRI modalities. This design allows for the effective fusion of complementary features from multiple modalities, thereby enhancing segmentation performance by leveraging the valuable information contained in each modality. Meanwhile, the HACM reduces computational complexity using a pseudo-3D convolution approach. This approach breaks down 3D convolutions into components along the transverse and sagittal axes. Unlike traditional 2D convolutions, this method preserves essential spatial information across dimensions. It ensures accurate segmentation while maximizing efficiency by capitalizing on the varying focus of information in different spatial planes. This approach takes advantage of the varying information density in these dimensions, achieving a balance between accuracy and efficiency. Through extensive experiments on the BraTS2021 dataset, our proposed modality fusion-based network under limited resources (LFBTS) achieves dice scores of 0.925, 0.911, and 0.886 for whole tumor (WT), tumor core (TC), and enhanced tumor (ET), respectively. These results outperform state-of-the-art (SOTA) models and consistently demonstrate superiority over models developed in the preceding 2 years. This highlights the potential of our approach in advancing brain tumor segmentation and improving clinical decision-making, particularly in settings with limited resources.

To better automate the diagnosis of autism spectrum disorder (ASD) and improve diagnostic accuracy, a graph neural network via dual-atlas multi-feature learning (DML-GNN) model for ASD diagnosis is constructed based on the local feature information of brain atlas and the global feature information from the multi-modal data. First, DML-GNN constructs a dual-atlas feature extraction module to capture the initial features of each subject. Second, it combines K-nearest-neighbor graphs, graph pooling, graph convolution (GCN) and graph channel attention (GCA) to construct a local feature learning module. This module extracts deep features for each subject and eliminate redundant features, and further fuses multi-atlases features efficiently. Third, DML-GNN constructs a global feature learning module by combining the non-imaging information of fMRI data and graph isomorphism network (GINConv), which combines the information of multi-modal data to construct comprehensive multi-graph features and learns node embeddings using GINConv. Finally, multi-layer perceptron (MLP) is used to obtain the final ASD diagnosis results. Compared with recent algorithms for ASD diagnosis on the public data set-Autism Brain Imaging Data Exchange I (ABIDE I), our method demonstrated superior performance, underscoring its potential as an effective tool.

Accurate liver tumor detection and segmentation are essential for disease diagnosis and treatment planning. While traditional methods rely on pixel-level mask annotations in fully supervised training, weakly supervised techniques are gaining attention due to their reduced annotation requirements. In this study, we propose an enhanced version of BoxInst, called Enhanced BoxInst, which incorporates two key innovations: the position activation (PA) Module and the progressive mask generation (PMG) Module. The PA Module utilizes a Spatial Awareness (SA) Block to accurately locate tumor regions and encodes the location information to the segmentation branch through the Spatial Interaction Encoding (SIE) mechanism, thereby achieving cross-spatial feature interaction and ultimately improving the segmentation accuracy of liver tumors. Additionally, the PMG Module employs a feature decomposition scheme to refine tumor masks progressively from coarse to fine, accurately restoring the overall layout and boundary details of the tumor mask. Extensive experiments on the LiTS, AMU-Liver, and 3DIRCADb datasets demonstrate that Enhanced BoxInst outperforms existing methods in liver tumor instance segmentation. These results highlight the potential of our approach for practical use in medical image analysis, especially when only box annotations are available. The code is available at https://github.com/ssli23/Enhanced_BoxInst.

The incidence and mortality of liver cancer in China are not optimistic. Early diagnosis and treatment have become the urgent means to solve this situation. To develop an improved radiomics model for the classification of hepatic nodules based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). The DCE-MRI images of 30 hepatitis, 30 cirrhotic nodules (CN), 30 dysplastic nodules (DN), and 30 hepatocellular carcinoma (HCC) patients were retrospectively and randomly divided into training and testing datasets in a 7:3 ratio. Firstly, the radiomics features of lesions were extracted by using feature extractor module based on Pyradiomics, from which optimal features were selected by least absolute shrinkage and selection operator (LASSO). Then, the improved whale optimization algorithm (WOA) with Tent mapping, Adaptive weight, and Levy flight (TALWOA) was used for parameter optimization of support vector machines (SVM). Finally, TALWOA-SVM was employed for the four-class classification of hepatic nodules. Receiver operating characteristic (ROC) curves, area under curve (AUC), and F1-score were used to evaluate the performance of the TALWOA-SVM model. Forty-four most informative features were selected from 851 features to train the SVM classifier. Compared with the standard whale algorithm and other optimization algorithms, the optimized model proposed in this paper has highest classification accuracy (81.315%), the ROC of each category being closer to the top left corner with AUC were 0.9378 (95% CI: 0.893–0.981), 0.9223 (95% CI: 0.873–0.971), 0.9794 (0.958–1.000), 0.9872 (0.971–1.000). The model proposed in this study can better classify hepatic nodules in different periods, and is expected to provide help for the early diagnosis of liver cancer.

Accurately segmenting gastrointestinal (GI) disease regions from Wireless Capsule Endoscopy images is essential for clinical diagnosis and survival prediction. However, challenges arise due to similar intensity distributions, variable lesion shapes, and fuzzy boundaries. In this paper, we propose MLFE-UNet, an advanced fusion of CNN-based transformers with UNet. Both the encoder and decoder utilize a multi-level feature extraction (MLFA) CNN-Transformer-based module. This module extracts features from the input data, considering both global dependencies and local information. Furthermore, we introduce a multi-level spatial attention (MLSA) block that functions as the bottleneck. It enhances the network's ability to handle complex structures and overlapping regions in feature maps. The MLSA block captures multiscale dependencies of tokens from the channel perspective and transmits them to the decoding path. A contextual feature stabilization block follows each transition to emulate lesion zones and facilitate segmentation guidelines at each phase. To address high-level semantic information, we incorporate a computationally efficient spatial channel attention block. This is followed by a stabilization block in the skip connections, ensuring global interaction and highlighting important semantic features from the encoder to the decoder. To evaluate the performance of our proposed MLFE-UNet, we selected common GI diseases, specifically bleeding and polyps. The dice coefficient scores obtained by MLFE-UNet on the MICCAI 2017 (Red lesion) and CVC-ClinicalDB data sets are 92.34% and 88.37%, respectively.