Age-related macular degeneration (AMD) is a leading cause of visual impairment in middle-aged and elderly populations. Existing diagnostic methods are often inefficient and subjective. To achieve accurate and efficient diagnosis of AMD lesions, this paper designs BELD-Net, an efficient segmentation network based on a large visual model. The model employs DINOv2 as its backbone to capture long-range and cross-scale features of AMD lesions through large-scale self-supervised pre-training. Combined with the query-based segmentation mechanism of Mask2Former, it significantly improves generalization in few-shot scenarios. BELD-Net incorporates LoRA modules into the Transformer layers of DINOv2 for low-rank adaptation, reducing computational cost and enhancing the model’s adaptability to fundus images while avoiding overfitting. Additionally, a Multi-scale Boundary Enhancement Module (MBEM) is designed using heterogeneous convolutions and an adaptive squeeze-and-excitation (SE) mechanism to address blurred and variable-sized lesion boundaries, thereby improving sensitivity to low-contrast edges. Experimental results demonstrate that BELD-Net achieves outstanding performance in automatic multi-lesion segmentation of AMD, with an mDice value of 59.61

With the continuous improvement of processes by textile manufacturers, the requirements for pilling of fabric materials are also increasing. At present, in the textile industry, the rating method for fabric pilling is to compare the sample with the standard sample in a specific environment to determine the grade of fabric pilling. However, this method is greatly influenced by subjective factors, and when facing a large number of samples, it consumes a large amount of manpower and material resources. Therefore, an objective, stable, and highly accurate rating method is needed. This paper proposes an end-to-end objective rating method based on a convolutional neural network model called DENet. The network employs a two-branch architecture that integrates an attention mechanism with multiscale convolution and connects large convolutional kernels across layers. To align with the subjective grading criteria for fabric pilling, this study simulates the pilling process of eight types of fabrics using two methods: the pilling box method and the circular trajectory pilling method. Consequently, eight fabric pilling image datasets were generated, covering knitted fabrics, woven fabrics, and nonwoven fabrics, including two patterned fabric types. The experiment results show that the accuracy of the rating method proposed in this paper is 97.95% for combined training on eight datasets, and the average accuracy is 98.07% for separate training on eight datasets, and the accuracy and generalization of the designed network model is high. The proposed system provides an efficient, objective, and stable solution for fabric pilling evaluation.

Glaucoma is a common ocular condition that can cause significant physiological distress and may lead to vision loss in advanced stages. In clinical practice, segmenting the optic cup and disc regions in fundus images provides key physiological indicators, such as the cup-to-disc ratio and optic nerve fiber layer thickness. These indicators are crucial for clinicians to make accurate diagnoses and informed treatment decisions. Therefore, this research is of significant clinical and societal relevance for the early detection and grading of glaucoma. This thesis introduces a novel network architecture, SMS-U-Net (SE Multi-resolution module Self-attention U-Net), based on the U-Net framework, for joint segmentation of the optic cup and disc in fundus images. The proposed approach scales features from input images and uses the SE (Squeeze-and-Excitation) module to adaptively extract features from each feature map, minimizing information loss. To preserve image details and capture features from multiple receptive fields before downsampling, the conventional convolutional sequence in the encoding path is replaced by a multi-resolution module employing 3 × 3, 5 × 5, and 7 × 7 convolutional kernels. Furthermore, a self-attention mechanism is integrated into the decoder to highlight critical features for segmentation. Experimental results show that the proposed SMS-U-Net method achieves an optic disc Dice coefficient of 0.985 and an IoU of 0.968, as well as an optic cup Dice coefficient of 0.919 and an IoU of 0.851, demonstrating its effectiveness.

Gastrointestinal stromal tumor (GIST) is the most common mesenchymal tumor of the gastrointestinal tract. GIST risk classification based on CT images is of great clinical significance. However, there are still some issues with the current methods in predicting the risk level of gastrointestinal stromal tumors. Existing 2D CNN methods using only a single CT image cannot fully characterize tumors; 3D CNN methods require a large amount of labeled samples to achieve better performance and have significant computational requirements. To this end, we propose a novel Inter-Slice Attention-based Vision Transformer (ISA-ViT) method that automatically integrates the sequential information of CT images to learn the three-dimensional features of tumors for GIST risk classification. Specifically, we incorporate inter-slice attention mechanism into the Vision Transformer to learn the sequential information among CT slices, so that three-dimensional features can be learnt by our model. The extracted CT image information with clinical information of patients are then fused for GIST risk classification. The experimental results show that our model reaches 84% in terms of classification Accuracy (ACC) and 0.94 in terms of Area Under Curve (AUC). Compared with other baseline models, the proposed model achieves state-of-the-art performance results.

A metal fatigue damage model is established in this study by employing real-time strain monitoring to evaluate the damage state of metal materials. The fatigue life simulation, based on crystal plasticity finite element analysis, establishes the constitutive relationship between strain and damage before microcrack initiation in the low-cycle fatigue state of aerospace aluminum alloy. Subsequently, a comprehensive analysis of the strain–damage relationship is conducted under various stress conditions. Electron backscattering diffraction analysis (EBSD) is used to examine the fatigue damage state of the grooved specimen before initiating fatigue cracks at various stages. This analysis validates the metal fatigue damage model proposed in this paper and is based on strain monitoring, contributing to the enhanced confirmation of the model’s accuracy.

Diabetic macular edema (DME) is a typical fundus disease that can cause blindness in severe cases. The morphology of the inner limiting membrane (ILM) to the retinal pigment epithelium (RPE) layer in the retina and the macular edema (ME) area are important features for the diagnosis of DME. Doctors usually use non-invasive and high-resolution optical coherence tomography (OCT) to examine the fundus of the patient. However, manual diagnosis has low efficiency and strong subjectivity. Realizing the automatic segmentation of the ILM-RPE layer and ME is extremely important for the early diagnosis of DME. In this paper, the attention mechanism based on residual convolution module U-Net (RCU-Net) is proposed for the automatic segmentation of the retinal layer and cystoid edema lesions. Through the fusion of the residual structure and CBAM for feature extraction, the useful features in the channel and space are effectively strengthened, and the network can better learn different levels of information. The proposed network is combined with the Lovasz-softmax loss, which can better target the correlation between targets to obtain the optimal segmentation model during training. Finally, this paper compares the proposed method with several other segmentation methods. The experimental results show that the MIoU$$ MIoU $$ of the method in this model reaches 88.595%, and the Accuracy$$ Accuracy $$ reaches 99.171%. The RCU-Net proposed in this paper is used to segment the ILM-RPE layer and ME region in the retina OCT B-scan images, and its overall performance is better than other networks.

The nucleus carries a wealth of genetic information that controls and regulates the characteristics and functions of cells. Its morphology and distribution are highly crucial for the differentiation and grading of tumors. However, with the widespread adoption of medical imaging technology, the task of handling massive cell images has become challenging and time-consuming, demanding specialized knowledge and expertise from medical professionals. The traditional nuclear image segmentation methods require expensive time consuming and exhibit limited adaptability to different environments. To address these problems, we propose a novel multi-scale TransUnet combined with Convolutional Block Attention Module (CBAM) method for nuclear image segmentation. Firstly, in the downsampling phase of the encoder, we introduce dilated convolution units with varying dilation rates. This enables the model to integrate multi-scale feature information, thereby expanding the receptive field during downsampling to extract cell nucleus information more comprehensively. In addition, to enhance the ability of target feature extraction, CBAM is added to the up-sampling of the decoder to sequentially acquire inter-channel dependencies and spatial pixel-level relationships. The experimental results show that the Dice coefficient of our model on the MoNuSeg dataset reaches 86.3%, and the crossover ratio of IoU reaches 84.2%. Compared with the second-best models (TransUnet), the performance of our model is improved by 1.22% and 0.45% in terms of IoU and Dice similarity coefficient, respectively.

Tuberculosis is a common infectious disease in the world. Tuberculosis cavities are common and an important imaging signs in tuberculosis. Accurate segmentation of tuberculosis cavities has practical significance for indicating the activity of lesions and guiding clinical treatment. However, this task faces challenges such as blurred boundaries, irregular shapes, different location and size of lesions and similar structures on computed tomography (CT) to other lung diseases or tissues. To overcome these problems, we propose a novel RNN-combined graph convolutional network (R2GCN) method, which integrates the bidirectional recurrent network (BRN) and graph convolution network (GCN) modules. First, feature extraction is performed on the input image by VGG-16 or ResNet-50 to obtain the feature map. The feature map is then used as the input of the two modules. On the one hand, we adopt the BRN to retrieve contextual information from the feature map. On the other hand, we take the vector for each location in the feature map as input nodes and utilize GCN to extract node topology information. Finally, two types of features obtained fuse together. Our strategy can not only make full use of node correlations and differences, but also obtain more precise segmentation boundaries. Extensive experiments on CT images of cavitary patients with tuberculosis show that our proposed method achieves the best segmentation accuracy than compared segmentation methods. Our method can be used for the diagnosis of tuberculosis cavity and the evaluation of tuberculosis cavity treatment.

Anti-pilling performance is an important indicator of fabric. In order to overcome the inefficiency and poor consistency of subjective pilling rating in the industry, we propose an objective rating method based on a convolutional neural network (CNN). We begin by establishing a pilling image dataset of four different fabrics and three different hairball shapes, including knitted fabric and nonwoven fabric. Next, we use a SONet rating system model. The model consists of two branches, S branch and O branch. The S branch extracts the hairball feature of the pilling image through an attention mechanism, while the O branch extracts the hairball feature and fabric texture feature by factorizing the mixed feature maps according to frequency. The results show that the rating accuracy of the proposed SONet model reaches 97.70%. Finally, we demonstrate the reliability of the SONet model in the objective evaluation of fabric pilling using a feature map, heat map and class probability scatter plot.

PURPOSE:The segmentation of retinal layers and fluid lesions on retinal optical coherence tomography (OCT) images is an important component of screening and diagnosing retinopathy in clinical ophthalmic treatment. We designed a novel network for accurate segmentation of the seven tissue layers of the retina and lesion areas of diabetic macular edema (DME), which can assist doctors to quantitatively analyze the disease. METHODS:In this article, we propose the Retinal Layer Macular Edema Network (RLMENet) model to achieve end-to-end joint segmentation of retinal layers and fluids. The network employs dense multi-scale attention to enhance the extraction of retinal layer and fluid detail information and achieve efficient long-range modeling, which improves the receptive field and obtains multi-scale features. As the more complex decoder part is designed, which integrates more low-level feature information on the decoder side, more features are extracted to gradually restore the resolution of the feature map and improve the segmentation accuracy. RESULTS:We used part of the OCT2017 dataset to train and verify the model to divide the data into a training set, validation set, and test set and set it to a 7:2:1 ratio. We evaluated our method on the ISIC2017 dataset. Experimental results showed that the RLMENet model designed in this work can accurately segment seven retinal tissue layers and DME lesions on the retinal OCT dataset. Finally, the MIoU value in the test set reached 86.55%. The model can be extended to other medical image segmentation datasets to achieve better segmentation performance. CONCLUSIONS:The proposed method was superior to the existing segmentation methods, achieved a more refined segmentation effect, and provided an auxiliary analysis tool for clinical diagnosis and treatment.

Crops are attacked by a variety of pests and diseases during their growth. Different pests have different control measures, and being able to accurately identify pests has become the key to pest control. Traditional methods have relatively low accuracy in pest identification due to the complexity of their algorithms and their susceptibility to environmental interference. This paper proposes an end-to-end pest identification network that combines deep learning and hyperspectral imaging technology. This method can identify common pests for the purposes of effective pest control. Noise and redundant information in the hyperspectral image (HSI) spectral space are treated by one-dimensional convolution and the attention mechanism between spectral channels to design a spectral feature extraction module for the efficient use of spectral information. The three-dimensional convolution branch structure of different resolutions in parallel is used as the HSI feature extractor to secure rich spectral–spatial information. The output feature map maintains high resolution throughout its usage. To further enhance the feature extraction capabilities of the network, an adaptive spectral–spatial feature extraction module is inserted into each branch to dynamically weight different information, thereby reducing the HSI’s undue influence. A hyperspectral imaging system was used to collect pest HSI, and a dataset containing nine kinds of common pests was constructed accordingly. The above method is used to test on this dataset, and the experimental results prove that this method has higher pest identification accuracy and is more suitable for pest identification tasks than other methods.

In the objective evaluation of wool knitted fabric pilling, the feature extraction step is a key factor affecting performance. In this paper, we proposed a double-branch deep cross-level fusion convolutional neural network (D-DCFNet) to improve feature selection. First, a cross-level fusion module (CLF module), based on a Fire module in SqueezeNet, was created to improve the feature extraction capability of a single module. Then, we designed a double-branch structure D-DCFNet. One branch consists of a CLF module, the core feature extraction module, and the other branch consists of a Fire module. Next, the features extracted from the two branches were fused together. Finally, the model trained by D-DCFNet was used to classify the knitting pilling data set to evaluate the robustness of the model. Experiments showed that D-DCFNet's rating accuracy for woolen knitted fabrics and semi-worsted knitted fabrics is 99.35% and 99.02%, respectively, when the model size is only 5.77 M.

Alzheimer's disease (AD) is a complex neurodegenerative disease involving a variety of pathogenic factors.So far,the cause of Alzheimer's disease is not clear,the course of the disease is irreversible,and there is no cure.Its early diagnosis and treatment have always been the focus of attention.The neuroimaging data of subjects has an important auxiliary role in the diagnosis of this disease,and the combination of multimodal data can further improve the diagnostic effect.At present,the multimodal data representation learning of the disease has gradually become an emerging research field,which has attracted wide attention from researchers.An autoencoder based multimodal representation learning method for Alzheimer's disease diagnosis is proposed.Firstly,the multimodal data are initially fused to obtain the primary common representation.Then,it is input into the autoencoder network to learn the final common representation in latent space.Finally,the common representation in latent space is classified to obtain the disease result.The proposed method,which achieves the best diagnostic results compared with comparison algorithms,has an accuracy of 88.9% in the classification of AD and healthy subjects in the ADNI dataset.Extensive experimental results verify its effectiveness.

At present, there are more and more people suffering from retinal diseases. Doctors can diagnose and prevent eye diseases by observing the changes in the thickness of the retinal layer in OCT images. Due to the low contrast of the retinal layer boundary of the OCT image, manual segmentation is time-consuming and laborious. Moreover, most of the current automatic retinal layer segmentation methods are based on traditional methods and the segmentation result is not good. Therefore, in this paper, we proposed an end-to-end automatic retinal layer segmentation method based on deep learning, called DA-PSPNet, which can accurately segment seven retinal layers in OCT images. DA-PSPNet integrates a dual attention mechanism based on the PSPNet network, aiming to extract richer layer boundary information. It merges features of various levels to aggregate contextual information in different regions. The experimental results show that the proposed method achieves better performance in several evaluation indexes compared with the other four mainstream segmentation networks.

Automatic fabric defect detection systems improve the quality of textile production across the industry. To make these automatic systems accessible to smaller businesses, one potential solution is to use limited memory capacity chips that can be used with hardware platforms with limited resources. That is to say, the fabric defect detection algorithm must ensure high detection accuracy while maintaining a low computational cost. Therefore, we propose a wide-and-light network structure based on Faster R-CNN for detecting common fabric defects. We enhance the feature extraction capability of the feature extraction network by designing a dilated convolution module. In a dilated convolution module, a multi-scale convolution kernel is used to adapt to defects of different sizes. Dilated convolutions can increase receptive fields without increasing the number of parameters used. Therefore, we replace a subset of ordinary convolutions with dilated convolutions to learn target features and use convolution kernel decomposition and bottleneck methods to simplify the feature extraction networks. Then, high-level semantic features are fused with bottom-level detail features (via skip-connection) to obtain multi-scale fusion features. Finally, a series of anchor frames (of different sizes) is designed to suit multi-scale fabric defect detection. Experiments show that compared with various mainstream target detection algorithms, our proposed algorithm can improve the accuracy of fabric defect detection and reduce the size of the model.

Pitch length and surface braiding angle are two important parameters of braided composite preforms. In this paper, a method based on Faster R-CNN is proposed to measure the two parameters. First, after image acquisition, a fabric image database including initial cropped images, augmented images, and target images is established. Then, the target images are classified into four categories according to the gray change characteristics. Third, a Faster R-CNN fabric detection model is trained on the fabric image database. Fourth, targets are detected by the trained network, and corners are detected based on the detected targets. Finally, pitch lengths and surface braiding angles are measured based on the detected corners. Experimental results show that the proposed method achieves the automatic measurement of pitch lengths and surface braiding angles of 2D and 3D braided composite preforms with high accuracy.

Diabetic macular edema (DME) is the main cause of visual impairment in diabetic patients. Early detection of DME will significantly reduce the risk of vision loss for the patients. According to the clinical DME grading standard, the positional relationship between Hard Exudates (HEs) and macular center is an important basis for DME grading. Accurate DME grading is thus predicated on properly locating the macular center and segmenting HEs. HEI-MED and E-ophtha EX data sets were tested by the proposed DME grading method, reaching an average accuracy of 94.4% and 87%, respectively. The proposed method was also tested by comparison against other commonly used methods as per its potential to assist doctors in initially screening DME; it was found to not only improve the efficiency of DME detection, but also to save Optical Coherence Tomography medical resources over the other methods tested.

眼底视网膜是唯一可用肉眼直接并集中观察到动脉、静脉与毛细血管的部位,因而眼底图像成为医生诊断眼底疾病及糖尿病、高血压、高血脂等疾病的重要依据.高质量的眼底图像是医生对眼底疾病患者进行病情诊断与治疗的前提.根据眼底相机采集到的视网膜图像中眼底结构清晰度、图像对比度等条件对眼底图像质量进行分类成为一个既具有研究价值又极具挑战性的难点问题.首先简述了眼底图像质量分类的研究意义和实用价值,回顾了其发展历史;然后介绍了方法分类、每类方法的基本思想并梳理了各类方法中代表性算法及其特点;之后针对用于眼底质量分类的数据集,分析比较了主要眼底图像质量分类方法的性能.分析表明,传统方法中依据眼底结构特征判断视网膜图像质量相较于通用图像质量参数更加客观,而随着神经网络与机器学习的出现,在大数据驱动下,基于卷积神经网络的质量分类方法在准确率与鲁棒性方面性能更佳.最后对眼底图像质量分类未来的发展趋势进行展望.

Doctors can diagnose retinal arteriosclerosis by observing the fundus image. However, the workload is large. This process is inefficient and accordingly unsuitable for the screening of retinal arteriosclerosis in large population. To address this issue, the current study proposes a retinal arteriosclerosis detection method on the basis of an improved VGG-16 network. First, the proposed method extracts multiscale features on the basis of the VGG-16 network by utilizing a dense block to enhance feature transfers and applying them to the segmentation of fundus arteriovenous vessels and arterial reflective bands. Test results reveal that the segmentation accuracy, sensitivity and specificity of the proposed method is 92.9%, 85.21%, and 93.40%, respectively. Subsequently, the effective region of the reflective band is screened, and a four-segment Gaussian model is utilized to fit the grayscale of the cross section of arterial blood vessel and reflective band in the effective region. Finally, the proposed method quantitatively detects retinal arteriosclerosis and calculates the reflective parameters of arterial blood vessels and reflective bands, such as bandwidth and grayscale ratios. A patient with retinal arteriosclerosis can be determined on the basis of the reflective parameters. The average detection accuracy of retinal arteriosclerosis is 92.3%, which is in accordance with the requirements of large-scale retinal arteriosclerosis.