International Journal of Imaging Systems and Technology最新文献

The literature has widely described the interaction between cardiac complications and COVID-19. However, the diagnosis of cardiac complications caused by COVID-19 using Computed Tomography (CT) images remains a challenge due to the diverse clinical manifestations. To address this issue, this study proposes a novel configuration of Convolutional Neural Network (CNN) for detecting cardiac complications derived from COVID-19 using CT images. The main contribution of this work lies in the use of CNN techniques in combination with Long Short-Term Memory (LSTM) for cardiac complication detection. To explore two-class classification (COVID-19 without cardiac complications vs. COVID-19 with cardiac complications), 10 650 CT images were collected from COVID-19 patients with and without myocardial infarction, myocarditis, and arrhythmia. The information was annotated by two radiology specialists. A total of 0.926 was found to be the accuracy, 0.84 was the recall, 0.82 was the precision, 0.82 was the F₁-score, and 0.830 was the g-mean of the suggested model. These results show that the suggested approach can identify cardiac problems from COVID-19 in CT scans. Patients with COVID-19 may benefit from the proposed LSTM-CNN architecture's enhanced ability to identify cardiac problems.

A multiscale feature fusion of Efficient-Residual Net is proposed for classifying tumors on brain Magnetic resonance images with solid or cystic masses, inadequate borders, unpredictable cystic and necrotic regions, and variable heterogeneity. Therefore, in this research, Efficient-Residual Net is proposed by efficaciously amalgamating features of two Deep Convolution Neural Networks—ResNet50 and EffficientNetB0. The skip connections in ResNet50 have reduced the chances of vanishing gradient and overfitting problems considerably thus learning of a higher number of features from input MR images. In addition, EffficientNetB0 uses a compound scaling coefficient for uniformly scaling the dimensions of the network such as depth, width, and resolution. The hybrid model has improved classification results on brain tumors with similar morphology and is tested on three internet repository datasets, namely, Kaggle, BraTS 2018, BraTS 2021, and real-time dataset from Post Graduate Institute of Medical Education and Research (PGIMER), Chandigarh. It is observed that the proposed system delivers an overall accuracy of 96.40%, 97.59%, 97.75%, and 97.99% on the four datasets, respectively. The proposed hybrid methodology has given assuring results of 98%–99% of other statistical such parameters as precision, negatively predicted values, and F1 score. The cloud-based web page is also created using the Django framework in Python programming language for accurate prediction and classification of different types of brain tumors.

Increases in the prevalence of melanoma, the most lethal form of skin cancer, have been observed over the last few decades. However, the likelihood of a longer life span for individuals is considerably improved with early detection of this malignant illness. Even though the field of computer vision has attained a certain level of success, there is still a degree of ambiguity that represents an unresolved research challenge. In the initial phase of this study, the primary objective is to improve the information derived from input features by combining multiple deep models with the proposed Information-theoretic feature fusion method. Subsequently, in the second phase, the study aims to decrease the redundant and noisy information through down-sampling using the proposed entropy-controlled binary bat selection algorithm. The proposed methodology effectively maintains the integrity of the original feature space, resulting in the creation of highly distinctive feature information. In order to obtain the desired set of features, three contemporary deep models are employed via transfer learning: Inception-Resnet V2, DenseNet-201, and Nasnet Mobile. By combining feature fusion and selection techniques, we may effectively fuse a significant amount of information into the feature vector and subsequently remove any redundant feature information. The effectiveness of the proposed methodology is supported by an evaluation conducted on three well-known dermoscopic datasets, specifically $��{\mathrm{PH}}^2�$, ISIC-2016, and ISIC-2017. In order to validate the proposed approach, several performance indicators are taken into account, such as accuracy, sensitivity, specificity, false negative rate (FNR), false positive rate (FPR), and F1-score. The accuracies obtained for all datasets utilizing the proposed methodology are 99.05%, 96.26%, and 95.71%, respectively.

Radiotherapy treatment planning (RTP) requires both magnetic resonance (MR) and computed tomography (CT) modalities. However, conducting separate MR and CT scans for patients leads to misalignment, increased radiation exposure, and higher costs. To address these challenges and mitigate the limitations of supervised synthesis methods, we propose a novel unsupervised perceptual attention image synthesis model based on cycleGAN (PA-cycleGAN). The innovation of PA-cycleGAN lies in its model structure, which incorporates dynamic feature encoding and deep feature extraction to improve the understanding of image structure and contextual information. To ensure the visual authenticity of the synthetic images, we design a hybrid loss function that incorporates perceptual constraints using high-level features extracted by deep neural networks. Our PA-cycleGAN achieves notable results, with an average peak signal-to-noise ratio (PSNR) of 28.06, structural similarity (SSIM) of 0.95, and mean absolute error (MAE) of 46.90 on a pelvic dataset. Additionally, we validate the generalization of our method by conducting experiments on an additional head dataset. These experiments demonstrate that PA-cycleGAN consistently outperforms other state-of-the-art methods in both quantitative metrics and image synthesis quality.

The manifestations of early lung cancer in medical imaging often appear as pulmonary nodules, which can be classified as benign or malignant. In recent years, there has been a gradual application of deep learning-based computer-aided diagnosis technology to assist in the diagnosis of lung nodules. This study introduces a novel three-dimensional (3D) residual network called SSANet, which integrates split-based convolution, shuffle attention, and a novel activation function. The aim is to enhance the accuracy of distinguishing between benign and malignant lung nodules using convolutional neural networks (CNNs) and alleviate the burden on doctors when interpreting the images. To fully extract pulmonary nodule information from chest CT images, the original residual network is expanded into a 3D CNN structure. Additionally, a 3D split-based convolutional operation (SPConv) is designed and integrated into the feature extraction module to reduce redundancy in feature maps and improve network inference speed. In the SSABlock part of the proposed network, ACON (Activated or Not) function is also introduced. The proposed SSANet also incorporates an attention module to capture critical characteristics of lung nodules. During the training process, the PolyLoss function is utilized. Once SSANet generates the diagnosis result, a heatmap displays using Score-CAM is employed to evaluate whether the network accurately identifies the location of lung nodules. In the final test set, the proposed network achieves an accuracy of 89.13%, an F1-score of 84.85%, and a G-mean of 86.20%. These metrics represent improvements of 5.43%, 5.98%, and 4.09%, respectively, compared with the original base network. The experimental results align with those of previous studies on pulmonary nodule diagnosis networks, confirming the reliability and clinical applicability of the diagnostic outcomes.

Diabetic retinopathy (DR) poses a serious threat to vision, emphasising the need for early detection. Manual analysis of fundus images, though common, is error-prone and time-intensive. Existing automated diagnostic methods lack precision, particularly in the early stages of DR. This paper introduces the Soft Convolutional Block Attention Module-based Network (Soft-CBAMNet), a deep learning network designed for severity detection, which features Soft-CBAM attention to capture complex features from fundus images. The proposed network integrates both the convolutional block attention module (CBAM) and the soft-attention components, ensuring simultaneous processing of input features. Following this, attention maps undergo a max-pooling operation, and refined features are concatenated before passing through a dropout layer with a dropout rate of 50%. Experimental results on the APTOS dataset demonstrate the superior performance of Soft-CBAMNet, achieving an accuracy of 85.4% in multiclass DR grading. The proposed architecture has shown strong robustness and general feature learning capability, achieving a mean AUC of 0.81 on the IDRID dataset. Soft-CBAMNet's dynamic feature extraction capability across all classes is further justified by the inspection of intermediate feature maps. The model excels in identifying all stages of DR with increased precision, surpassing contemporary approaches. Soft-CBAMNet presents a significant advancement in DR diagnosis, offering improved accuracy and efficiency for timely intervention.

Medical image labeling requires specialized knowledge; hence, the solution to the challenge of medical image classification lies in efficiently utilizing the few labeled samples to create a high-performance model. Building a high-performance model requires a complicated convolutional neural network (CNN) model with numerous parameters to be trained which makes the test quite expensive. In this paper, we propose optimizing a lightweight deep learning model with only five convolutional layers using the particle swarm optimization (PSO) algorithm to find the best number of kernel filters for each convolutional layer. For colored red, green, and blue (RGB) images acquired from different data sources, we suggest using stain separation using color deconvolution and horizontal and vertical flipping to produce new versions that can concentrate the representation of the images on structures and patterns. To mitigate the effect of training with incorrectly or uncertainly labeled images, grades of disease could have small variances, we apply a second-pass training excluding uncertain data. With a small number of parameters and higher accuracy, the proposed lightweight deep learning model optimization (LDLMO) algorithm shows strong resilience and generalization ability compared with most recent research on four MedMNIST datasets (RetinaMNIST, BreastMNIST, DermMNIST, and OCTMNIST), Medical-MNIST, and brain tumor MRI datasets.

Skin cancer is a pressing global health issue, with high incidence and mortality rates. Convolutional neural network (CNN) models have been proven to be effective in improving performance in skin lesion image classification and reducing the medical burden. However, the inherent class imbalance in training data caused by the difficulty of collecting dermoscopy images leads to categorical overfitting, which still limits the effectiveness of these data-driven models in recognizing few-shot categories. To address this challenge, we propose a dynamic multi-output convolutional neural network (DMO-CNN) model that incorporates exit nodes into the standard CNN structure and includes feature refinement layers (FRLs) and an adaptive output scheduling (AOS) module. This model improves feature representation ability through multi-scale sub-feature maps and reduces the inter-layer dependencies during backpropagation. The FRLs ensure efficient and low-loss down-sampling, while the AOS module utilizes a trainable layer selection mechanism to refocus the model's attention on few-shot lesion categories. Additionally, a novel correction factor loss is introduced to supervise and promote AOS convergence. Our evaluation of the DMO-CNN model on the HAM10000 dataset demonstrates its effectiveness in multi-class skin lesion classification and its superior performance in recognizing few-shot categories. Despite utilizing a very simple VGG structure as the sole backbone structure, DMO-CNN achieved impressive performance of 0.885 in BACC and 0.983 in weighted AUC. These results are comparable to those of the ensemble model that won the ISIC 2018 challenge, highlighting the strong potential of DMO-CNN in dealing with few-shot skin lesion data.