Automatic Image Segmentation Method Research Articles

Detection of Road Crack Images Based on Multistage Feature Fusion and a Texture Awareness Method

Structural health monitoring for roads is an important task that supports inspection of transportation infrastructure. This paper explores deep learning techniques for crack detection in road images and proposes an automatic pixel-level semantic road crack image segmentation method based on a Swin transformer. This method employs Swin-T as the backbone network to extract feature information from crack images at various levels and utilizes the texture unit to extract the texture and edge characteristic information of cracks. The refinement attention module (RAM) and panoramic feature module (PFM) then merge these diverse features, ultimately refining the segmentation results. This method is called FetNet. We collect four public real-world datasets and conduct extensive experiments, comparing FetNet with various deep-learning methods. FetNet achieves the highest precision of 90.4%, a recall of 85.3%, an F1 score of 87.9%, and a mean intersection over union of 78.6% on the Crack500 dataset. The experimental results show that the FetNet approach surpasses other advanced models in terms of crack segmentation accuracy and exhibits excellent generalizability for use in complex scenes.

Multi-network approach for image segmentation in non-contrast enhanced cardiac 3D MRI of arrhythmic patients

Left atrial appendage (LAA) is the source of thrombi formation in more than 90% of strokes in patients with nonvalvular atrial fibrillation. Catheter-based LAA occlusion is being increasingly applied as a treatment strategy to prevent stroke. Anatomical complexity of LAA makes percutaneous occlusion commonly performed under transesophageal echocardiography (TEE) and X-ray (XR) guidance especially challenging. Image fusion techniques integrating 3D anatomical models derived from pre-procedural imaging into the live XR fluoroscopy can be applied to guide each step of the LAA closure. Cardiac magnetic resonance (CMR) imaging gains in importance for radiation-free evaluation of cardiac morphology as alternative to gold-standard TEE or computed tomography angiography (CTA). Manual delineation of cardiac structures from non-contrast enhanced CMR is, however, labor-intensive, tedious, and challenging due to the rather low contrast. Additionally, arrhythmia often impairs the image quality in ECG synchronized acquisitions causing blurring and motion artifacts. Thus, for cardiac segmentation in arrhythmic patients, there is a strong need for an automated image segmentation method. Deep learning-based methods have shown great promise in medical image analysis achieving superior performance in various imaging modalities and different clinical applications. Fully-convolutional neural networks (CNNs), especially U-Net, have become the method of choice for cardiac segmentation. In this paper, we propose an approach for automatic segmentation of cardiac structures from non-contrast enhanced CMR images of arrhythmic patients based on CNNs implemented in a multi-stage pipeline. Two-stage implementation allows subdividing the task into localization of the relevant cardiac structures and segmentation of these structures from the cropped sub-regions obtained from previous step leading to efficient and effective way of automated cardiac segmentation.

Petascale pipeline for precise alignment of images from serial section electron microscopy

The reconstruction of neural circuits from serial section electron microscopy (ssEM) images is being accelerated by automatic image segmentation methods. Segmentation accuracy is often limited by the preceding step of aligning 2D section images to create a 3D image stack. Precise and robust alignment in the presence of image artifacts is challenging, especially as datasets are attaining the petascale. We present a computational pipeline for aligning ssEM images with several key elements. Self-supervised convolutional nets are trained via metric learning to encode and align image pairs, and they are used to initialize iterative fine-tuning of alignment. A procedure called vector voting increases robustness to image artifacts or missing image data. For speedup the series is divided into blocks that are distributed to computational workers for alignment. The blocks are aligned to each other by composing transformations with decay, which achieves a global alignment without resorting to a time-consuming global optimization. We apply our pipeline to a whole fly brain dataset, and show improved accuracy relative to prior state of the art. We also demonstrate that our pipeline scales to a cubic millimeter of mouse visual cortex. Our pipeline is publicly available through two open source Python packages.

Deep learning image segmentation for the reliable porosity measurement of high-capacity Ni-based oxide cathode secondary particles

The optimization of geometrical pore control in high-capacity Ni-based cathode materials is required to enhance the cyclic performance of lithium-ion batteries. Enhanced porosity improves lithium-ion mobility by increasing the electrode–electrolyte contact area and reducing the number of ion diffusion pathways. However, excessive porosity can diminish capacity, thus necessitating optimizing pore distribution to compromise the trade-off relation. Accordingly, a statistically meaningful porosity estimation of electrode materials is required to engineer the local pore distribution inside the electrode particles. Conventional scanning electron microscopy (SEM) image-based porosity measurement can be used for this purpose. However, it is labor-intensive and subjected to human bias for low-contrast pore images, thereby potentially lowering measurement accuracy. To mitigate these difficulties, we propose an automated image segmentation method for the reliable porosity measurement of cathode materials using deep convolutional neural networks specifically trained for the analysis of porous cathode materials. Combined with the preprocessed SEM image datasets, the model trained for 100 epochs exhibits an accuracy of > 97% for feature segmentation with regard to pore detection on the input datasets. This automated method considerably reduces manual effort and human bias related to the digitization of pore features in serial section SEM image datasets used in 3D electron tomography.Graphical abstract

GRAF: Gap region aware framework for Al-Si alloy microscopic image segmentation

The shape of the primary silicon in Al-Si alloys is one of the main factors determining the properties of the alloy, which makes the microscopic image of Al-Si alloys become a crucial reference material in the study of the alloy. Accurate segmentation of silicon particles in Al-Si alloy micrographs is a necessary prerequisite for their quantitative analysis. The traditional manual annotation method requires high time and labor costs and will be affected by subjective factors of the annotator. Although automatic image segmentation methods can avoid these problems, there are silicon particles of different shapes and sizes in the Al-Si alloy microscopic image, and it is hard to separate adjacent silicon particles. These characteristics bring the automatic segmentation method severe challenge. Therefore, to address the issue of deviation in the annotation range when there is agglomeration of primary silicon particles in the microstructure images of Al-Si alloys, We propose a Residual Position-Aware Attention Network (RPAA-Net) for the segmentation of primary silicon particles. The RPAA-Net method enhances segmentation accuracy and robustness by incorporating a residual position-aware attention mechanism, enabling the network to better distinguish silicon particles from the surrounding non-silicon regions. To address the challenge of separating adjacent silicon particles in the image, we introduce a novel Gap Region Weighting Strategy (GRWS). GRWS enhances the model's focus on the gap regions between adjacent objects by applying additional weights during training. By combining RPAA-Net with GRWS, we construct a Gap Region Awareness Framework (GRAF) and experiment with different weights. We find that the weight with the most significant impact is 10, leading to the framework being named GRAF-10. Finally, we apply GRAF-10 to the Aluminum-Silicon Alloy Microstructure (ASAM) dataset. The results show that GRAF-10 outperforms the current state-of-the-art methods, improving the object-level Dice index(Dobj) by 1.56%. Additionally, it reduces the Merge Error (ME) metric by 6.74 and the Variation of Information (VI) metric by 5.43, effectively segmenting aluminum–silicon alloy microstructure images.

Automatic cochlear multimodal 3D image segmentation and analysis using atlas–model-based method

Objectives: To propose an automated fast cochlear segmentation, length, and volume estimation method from clinical 3D multimodal images which has a potential role in the choice of cochlear implant type, surgery planning, and robotic surgeries. Methods: Two datasets from different countries were used. These datasets include 219 clinical 3D images of cochlea from 3 modalities: CT, CBCT, and MR. The datasets include different ages, genders, and types of cochlear implants. We propose an atlas–model-based method for cochlear segmentation and measurement based on high-resolution μCT model and A-value. The method was evaluated using 3D landmarks located by two experts. Results: The average error was mm and the average time required to process an image was seconds (P<0.001). The volume of the cochlea ranged from 73.96 mm to 106.97 mm, the cochlear length ranged from 36.69 to 45.91 mm at the lateral wall and from 29.12 to 39.05 mm at the organ of Corti. Discussion: We propose a method that produces nine different automated measurements of the cochlea: volume of scala tympani, volume of scala vestibuli, central lengths of the two scalae, the scala tympani lateral wall length, and the organ of Corti length in addition to three measurements related to A-value. Conclusion: This automatic cochlear image segmentation and analysis method can help clinician process multimodal cochlear images in approximately 5 seconds using a simple computer. The proposed method is publicly available for free download as an extension for 3D Slicer software.

Energy minimization segmentation model based on MRI images.

Medical image segmentation is an important tool for doctors to accurately analyze the volume of brain tissue and lesions, which is important for the correct diagnosis of brain diseases. However, manual image segmentation methods are time-consuming, subjective and lack of repeatability, it needs to develop automatic and reliable methods for image segmentation. Magnetic Resonance Imaging (MRI), a non-invasive imaging technique, is commonly used to detect, characterize and quantify tissues and lesions in the brain. Partial volume effect, gray scale in homogeneity, and lesions presents a great challenge for automatic medical image segmentation methods. So, the paper is dedicated to address the impact of partial volume effect and multiple sclerosis lesions on the segmentation accuracy in MRI. The objective function of the improved model and the post-processing method of lesion filling are researched based on the fuzzy clustering space and energy model. In particular, an energy-minimized segmentation algorithm is proposed. Through experimental verification, the AR-FCM algorithm can better overcome the problem of low segmentation accuracy of the RFCM algorithm for tissue boundary voxels and improve the segmentation accuracy of this algorithm. Meanwhile, a multi-channel input energy-minimization segmentation method with lesion filling and anatomical mapping is further proposed. The feasibility of the lesion filling strategy using post-processing can be confirmed and the segmentation accuracy is increased by comparison experiments.

Automatic Segmentation and Classification for Antinuclear Antibody Images Based on Deep Learning.

Antinuclear antibodies (ANAs) testing is the main serological diagnosis screening test for autoimmune diseases. ANAs testing is conducted principally by the indirect immunofluorescence (IIF) on human epithelial cell-substrate (HEp-2) protocol. However, due to its high variability and human subjectivity, there is an insistent need to develop an efficient method for automatic image segmentation and classification. This article develops an automatic segmentation and classification framework based on artificial intelligence (AI) on the ANA images. The Otsu thresholding method and watershed segmentation algorithm are adopted to segment IIF images of cells. Moreover, multiple texture features such as scale-invariant feature transform (SIFT), local binary pattern (LBP), cooccurrence among adjacent LBPs (CoALBP), and rotation invariant cooccurrence among adjacent LBPs (RIC-LBP) are utilized. Firstly, this article adopts traditional machine learning methods such as support vector machine (SVM), k-nearest neighbor algorithm (KNN), and random forest (RF) and then uses ensemble classifier (ECLF) combined with soft voting rules to merge these machine learning methods for classification. The deep learning method InceptionResNetV2 is also utilized to train on the classification of cell images. Eventually, the best accuracy of 0.9269 on the Changsha dataset and 0.9635 on the ICPR 2016 dataset for the traditional methods is obtained by a combination of SIFT and RIC-LBP with the ECLF classifier, and the best accuracy obtained by the InceptionResNetV2 is 0.9465 and 0.9836 separately, which outperforms other schemes.

TransU²-Net: An Effective Medical Image Segmentation Framework Based on Transformer and U²-Net.

In the past few years, U-Net based U-shaped architecture and skip-connections have made incredible progress in the field of medical image segmentation. U2-Net achieves good performance in computer vision. However, in the medical image segmentation task, U2-Net with over nesting is easy to overfit. A 2D network structure TransU2-Net combining transformer and a lighter weight U2-Net is proposed for automatic segmentation of brain tumor magnetic resonance image (MRI). The light-weight U2-Net architecture not only obtains multi-scale information but also reduces redundant feature extraction. Meanwhile, the transformer block embedded in the stacked convolutional layer obtains more global information; the transformer with skip-connection enhances spatial domain information representation. A new multi-scale feature map fusion strategy as a postprocessing method was proposed for better fusing high and low-dimensional spatial information. Our proposed model TransU2-Net achieves better segmentation results, on the BraTS2021 dataset, our method achieves an average dice coefficient of 88.17%; Evaluation on the publicly available MSD dataset, we perform tumor evaluation, we achieve a dice coefficient of 74.69%; in addition to comparing the TransU2-Net results are compared with previously proposed 2D segmentation methods. We propose an automatic medical image segmentation method combining transformers and U2-Net, which has good performance and is of clinical importance. The experimental results show that the proposed method outperforms other 2D medical image segmentation methods. Clinical Translation Statement: We use the BarTS2021 dataset and the MSD dataset which are publicly available databases. All experiments in this paper are in accordance with medical ethics.

Dual-Tree Complex Wavelet Input Transform for Cyst Segmentation in OCT Images Based on a Deep Learning Framework

Optical coherence tomography (OCT) represents a non-invasive, high-resolution cross-sectional imaging modality. Macular edema is the swelling of the macular region. Segmentation of fluid or cyst regions in OCT images is essential, to provide useful information for clinicians and prevent visual impairment. However, manual segmentation of fluid regions is a time-consuming and subjective procedure. Traditional and off-the-shelf deep learning methods fail to extract the exact location of the boundaries under complicated conditions, such as with high noise levels and blurred edges. Therefore, developing a tailored automatic image segmentation method that exhibits good numerical and visual performance is essential for clinical application. The dual-tree complex wavelet transform (DTCWT) can extract rich information from different orientations of image boundaries and extract details that improve OCT fluid semantic segmentation results in difficult conditions. This paper presents a comparative study of using DTCWT subbands in the segmentation of fluids. To the best of our knowledge, no previous studies have focused on the various combinations of wavelet transforms and the role of each subband in OCT cyst segmentation. In this paper, we propose a semantic segmentation composite architecture based on a novel U-net and information from DTCWT subbands. We compare different combination schemes, to take advantage of hidden information in the subbands, and demonstrate the performance of the methods under original and noise-added conditions. Dice score, Jaccard index, and qualitative results are used to assess the performance of the subbands. The combination of subbands yielded high Dice and Jaccard values, outperforming the other methods, especially in the presence of a high level of noise.

Automated image segmentation of 3D printed fibrous composite micro-structures using a neural network

A new, automated image segmentation method is presented that effectively identifies the micro-structural objects (fibre, air void, matrix) of 3D printed fibre-reinforced materials using a deep convolutional neural network. The method creates training data from a physical specimen composed of a single, straight fibre embedded in a cementitious matrix with air voids. The specific micro-structure of this strain-hardening cementitious composite (SHCC) is obtained from X-ray micro-computed tomography scanning, after which the 3D ground truth mask of the sample is constructed by connecting each voxel of a scanned image to the corresponding micro-structural object. The neural network is trained to identify fibres oriented in arbitrary directions through the application of a data augmentation procedure, which eliminates the time-consuming task of a human expert to manually annotate these data. The predictive capability of the methodology is demonstrated via the analysis of a practical SHCC developed for 3D concrete printing, showing that the automated segmentation method is well capable of adequately identifying complex micro-structures with arbitrarily distributed and oriented fibres. Although the focus of the current study is on SHCC materials, the proposed methodology can also be applied to other fibre-reinforced materials, such as fibre-reinforced plastics. The micro-structures identified by the image segmentation method may serve as input for dedicated finite element models that allow for computing their mechanical behaviour as a function of the micro-structural composition.

Using Convolutional Neural Networks for Segmentation of Multiple Sclerosis Lesions in 3D Magnetic Resonance Imaging

Magnetic Resonance Imaging to detect its lesions is used to diagnose multiple sclerosis. Experts usually perform this detection process manually, but there is interest in automating it to speed up the diagnosis and monitoring of this disease. A variety of automatic image segmentation methods have been proposed to quickly detect these lesions. A Gaussian Mixture Model is first constructed to identify outliers in each image. Then, using a set of rules based on expert knowledge of multiple sclerosis lesions, those outliers of the model that do not match the lesions' characteristics are discarded. Furthermore, segmented lesions usually correspond to gray matter-rich brain regions. In some cases, false positives can be detected, but the rules used cannot eliminate all errors without jeopardizing the segmentation’s quality. The second method involves training a convolutional neural network (CNN) that can segment lesions based on a set of training images. This technique can learn a set of filters that, when applied to small sections of an image called “patches,” produce a set of characteristics that can be used to classify each voxel of the image as a lesion or healthy tissue. On the other hand, the results show that the networks are capable of producing results in the worked database comparable to those produced by the algorithms in the literature.

Comprehensive Study for Breast Cancer Using Deep Learning and Traditional Machine Learning

Comprehensive Study for Breast Cancer Using Deep Learning and Traditional Machine Learning

Deep active contours using locally controlled distance vector flow

Active contours model (ACM) has been extensively used in computer vision and image processing. In recent studies, convolutional neural networks (CNNs) have been combined with ACM replacing the user in the process of contour evolution and image segmentation to eliminate limitations associated with ACM dependence on energy functional parameters and initialization. However, prior studies did not aim for automatic initialization, which is addressed in this article. In addition to manual initialization, current methods are highly sensitive to the initial location and fail to delineate borders accurately. We propose a fully automatic image segmentation method to address problems of manual initialization, insufficient capture range, and poor convergence to boundaries, in addition to the problem of assignment of energy functional parameters. We train two CNNs, one of which generating ACM weighting parameters and the other generating a ground truth mask to extract distance transform (DT) and an initialization circle. DT is used to form a vector field pointing from each pixel of the image towards the closest ground truth boundary point. Vector magnitudes are equal to the Euclidean distance between each pixel and the closest ground truth boundary point. We evaluate our method on four publicly available datasets, including two building instance segmentation datasets, i.e., Vaihingen and Bing huts, and two mammography image datasets, INBreast and DDSM-BCRP. Our approach achieves state-of-the-art results in mean Intersection over Union (mIoU), Dice similarity coefficient and Boundary F-score (BoundF) with the values of 92.33%, 92.44%, and 86.57% for Vaihingen dataset, and 87.12%, 86.86%, and 66.91% for Bing huts dataset. We obtained the Dice similarity coefficient values of 94.23% and 90.89% for the INBreast and DDSM-BCRP, respectively.

Image Segmentation Method for Maize Ear Using Self-defined Activation Function

A Unet model is built segment the different parts of maize ear. However, the accuracy does not meet the needs of application. In the paper, a new activation function named R_S is defined on the basis of the analysis and research on two commonly used activation functions ReLU and Swish. And the new activation function combined the advantages of these two functions and effectively avoided the problems of hard saturation, mean shift and complicated calculations. The U-Net deep learning model with R_S as the activation function, which is used for image segmentation of maize ears, not only had fast convergence speed and high accuracy, but also avoided local convergence effectively. Next, in order to determine the optimal value of adjustment factor, seven U-Net models are built. The adjustment factors are separately 1.7,1.3,1.0,0.8,0.6,0.3,0.05. Experiment results showed that the model could reach the best comprehensive performance when using the self-defined activation function on condition that the adjustment factor is 0.8. Finally, the optimal model is used as the classifier to build the automatic maize ear image segmentation method. The experiment results demonstrate that the accuracy of the model on test set is 94.34%.

Automatic segmentation of breast cancer histological images based on dual-path feature extraction network.

The traditional manual breast cancer diagnosis method of pathological images is time-consuming and labor-intensive, and it is easy to be misdiagnosed. Computer-aided diagnosis of WSIs gradually comes into people*s sight. However, the complexity of high-resolution breast cancer pathological images poses a great challenge to automatic diagnosis, and the existing algorithms are often difficult to balance the accuracy and efficiency. In order to solve these problems, this paper proposes an automatic image segmentation method based on dual-path feature extraction network for breast pathological WSIs, which has a good segmentation accuracy. Specifically, inspired by the concept of receptive fields in the human visual system, dilated convolutional networks are introduced to encode rich contextual information. Based on the channel attention mechanism, a feature attention module and a feature fusion module are proposed to effectively filter and combine the features. In addition, this method uses a light-weight backbone network and performs pre-processing on the data, which greatly reduces the computational complexity of the algorithm. Compared with the classic models, it has improved accuracy and efficiency and is highly competitive.

A Robust SVM Color-Based Food Segmentation Algorithm for the Production Process of a Traditional Carasau Bread

In this paper, we address the problem of automatic image segmentation methods applied to the partial automation of the production process of a traditional Sardinian flatbread called pane Carasau for assuring quality control. The study focuses on one of the most critical activities for obtaining an efficient degree of automation: the estimation of the size and shape of the bread sheets during the production phase, to study the shape variations undergone by the sheet depending on some environmental and production variables. The knowledge can thus be used to create a system capable of predicting the quality of the shape of the dough produced and empower the production process. We implemented an image acquisition system and created an efficient machine learning algorithm, based on support vector machines, for the segmentation and estimation of image measurements for Carasau bread. Experiments demonstrated that the method can successfully achieve accurate segmentation of bread sheets images, ensuring that the dimensions extracted are representative of the sheets coming from the production process. The algorithm proved to be fast and accurate in estimating the size of the bread sheets in various scenarios that occurred over a year of acquisitions. The maximum error committed by the algorithm is equal to the 2.2% of the pixel size in the worst scenario and to 1.2% elsewhere.

Deep Hybrid Convolutional Neural Network for Segmentation of Melanoma Skin Lesion.

Melanoma is a type of skin cancer that often leads to poor prognostic responses and survival rates. Melanoma usually develops in the limbs, including in fingers, palms, and the margins of the nails. When melanoma is detected early, surgical treatment may achieve a higher cure rate. The early diagnosis of melanoma depends on the manual segmentation of suspected lesions. However, manual segmentation can lead to problems, including misclassification and low efficiency. Therefore, it is essential to devise a method for automatic image segmentation that overcomes the aforementioned issues. In this study, an improved algorithm is proposed, termed EfficientUNet++, which is developed from the U-Net model. In EfficientUNet++, the pretrained EfficientNet model is added to the UNet++ model to accelerate segmentation process, leading to more reliable and precise results in skin cancer image segmentation. Two skin lesion datasets were used to compare the performance of the proposed EfficientUNet++ algorithm with other common models. In the PH2 dataset, EfficientUNet++ achieved a better Dice coefficient (93% vs. 76%–91%), Intersection over Union (IoU, 96% vs. 74%–95%), and loss value (30% vs. 44%–32%) compared with other models. In the International Skin Imaging Collaboration dataset, EfficientUNet++ obtained a similar Dice coefficient (96% vs. 94%–96%) but a better IoU (94% vs. 89%–93%) and loss value (11% vs. 13%–11%) than other models. In conclusion, the EfficientUNet++ model efficiently detects skin lesions by improving composite coefficients and structurally expanding the size of the convolution network. Moreover, the use of residual units deepens the network to further improve performance.

Automated image segmentation method to analyse skeletal muscle cross section in exercise-induced regenerating myofibers

Skeletal muscle is an adaptive tissue with the ability to regenerate in response to exercise training. Cross-sectional area (CSA) quantification, as a main parameter to assess muscle regeneration capability, is highly tedious and time-consuming, necessitating an accurate and automated approach to analysis. Although several excellent programs are available to automate analysis of muscle histology, they fail to efficiently and accurately measure CSA in regenerating myofibers in response to exercise training. Here, we have developed a novel fully-automated image segmentation method based on neutrosophic set algorithms to analyse whole skeletal muscle cross sections in exercise-induced regenerating myofibers, referred as MyoView, designed to obtain accurate fiber size and distribution measurements. MyoView provides relatively efficient, accurate, and reliable measurements for CSA quantification and detecting different myofibers, myonuclei and satellite cells in response to the post-exercise regenerating process. We showed that MyoView is comparable with manual quantification. We also showed that MyoView is more accurate and efficient to measure CSA in post-exercise regenerating myofibers as compared with Open-CSAM, MuscleJ, SMASH and MyoVision. Furthermore, we demonstrated that to obtain an accurate CSA quantification of exercise-induced regenerating myofibers, whole muscle cross-section analysis is an essential part, especially for the measurement of different fiber-types. We present MyoView as a new tool to quantify CSA, myonuclei and satellite cells in skeletal muscle from any experimental condition including exercise-induced regenerating myofibers.

Precise segmentation of non-enhanced computed tomography in patients with ischemic stroke based on multi-scale U-Net deep network model

Background and ObjectiveAcute ischemic stroke requires timely diagnosis and thrombolytic therapy, but it is difficult to locate and quantify the lesion site manually. The purpose of this study was to explore a more rapid and effective method for automatic image segmentation of acute ischemic stroke. MethodsThe image features of 30 stroke patients were segmented from non-enhanced computed tomography (CT) images using a multi-scale U-Net deep network model. The Dice loss function training model was used to counter the similar imbalance problem in the data. The difference was compared between manual segmentation and automatic segmentation. ResultsThe Dice similarity coefficient based on multi-scale convolution U-Net network segmentation was 0.86±0.04, higher than the Dice based on classic U-Net (0.81±0.07, P=0.001). The lesion contour of automatic segmentation based on multi-scale U-Net was very close to manual segmentation. The error of lesion area is 1.28±0.59 mm2, and the Pearson correlation coefficient was r=0.986 (P<0.01). The motion time of automatic segmentation is less than 20 ms. ConclusionsMulti-scale U-Net deep network model can effectively segment ischemic stroke lesions in non-enhanced CT and meet real-time clinical requirements.