Proposed Segmentation Network Research Articles

Deep learning-based cervical lesion segmentation in colposcopic images

Artificial intelligence assisted cancer detection has changed the ream of diagnosis precision. This study aims to propose a segmentation network using artificial intelligence for accurately segmenting the cervix region and acetowhite lesions in cervigram images, addressing the shortage of skilled colposcopists and streamlining the training process. A computational approach is employed to develop and train a deep learning model specifically tailored for cervix region and acetowhite lesion segmentation in cervigram images. A dataset acquired in collaboration with KIDWAI memorial cancer research institute is used for building the model. Cervigram images are collected for training and validation, and a deep learning architecture is constructed and trained using annotated datasets. The segmentation network based on efficientnet architecture and atrous spatial pyramid pooling is designed to accurately identify and delineate the target regions, with performance evaluation conducted using precision, accuracy, recall, dice score, and specificity metrics. The proposed segmentation network achieves a precision of 0.7387±0.1541, accuracy of 0.9291, recall of 0.7912±0.1439, dice score of 0.7431±0.1506, and specificity of 0.9589±0.0131, indicating its reliability and robustness in segmenting cervix regions and acetowhite lesions in cervigram images. This research demonstrates the feasibility and effectiveness of using artificial intelligence-based computational models for cervix region and acetowhite lesion segmentation in cervigram images. It provides a foundation for further investigations into classifying cervix malignancy using AI techniques, potentially enhancing early detection and treatment of cervical cancer while addressing the shortage of skilled professionals in the field

Craniomaxillofacial Bone Segmentation and Landmark Detection Using Semantic Segmentation Networks and an Unbiased Heatmap.

Craniomaxillofacial (CMF) surgery always relies on accurate preoperative planning to assist surgeons, and automatically generating bone structures and digitizing landmarks for CMF preoperative planning is crucial. Since the soft and hard tissues of the CMF regions possess complicated attachment, segmenting the CMF bones and detecting the CMF landmarks are challenging problems. In this study, we proposed a semantic segmentation network to segment the maxilla, mandible, zygoma, zygomatic arch, and frontal bones. Then, we obtained the minimum bounding box around the CMF bones. After cropping, we used the top-down heatmap landmark detection network, similar to the segmentation module, to identify 18 CMF landmarks from the cropping patch. In addition, an unbiased heatmap encoding method was proposed to generate actual landmark coordinates in the heatmap. To overcome quantization effects in the heatmap-based landmark detection networks, the distribution-prior coordinate representation of medical landmarks (DCRML) was proposed to utilize the prior distribution of the encoding heatmap, approximating the accurate landmark coordinates in heatmap decoding by Taylor's theorem. The encoding and decoding method can easily contribute to other existing landmark detection frameworks based on heatmaps; consequently, these approaches can readily benefit without changing model structure. We used prior segmentation knowledge to enhance the semantic information around the landmarks, increasing landmark detection accuracy. The proposed framework was evaluated by 100 healthy persons and 86 patients from multicenter cooperation. The mean Dice score of our proposed segmentation network achieved over 88 %; in particular, the mandible accuracy was approximately 95%. The mean error of landmarks was 1.84 ±1.32 mm.

Segmentation and thickness calculation of glomerular basement membrane using RADS-Net in glomerular microscopic images

Glomerular ultrastructure needs to be observed under transmission electron microscopy during pathological diagnosis of renal biopsies. Evaluating the glomerular basement membranes (GBM) is necessary for the renal pathological diagnosis. There is little work that performs well for full-flow automated analysis of GBM in whole-glomerular electron microscopy images, including GBM segmentation, location, and thickness calculation. To address this problem, we propose a network architecture, RADS-Net, whose segmentation module combines the advantages of ViT and CNN to achieve better performance in GBM and electron dense deposit (EDD) segmentation tasks. It also contains an uncertainty evaluation module and a location module to exclude unsuitable regions for measurement and extract accurate GBM contours. We propose a GBM thickness calculation method based on statistical learning method, skeletonization algorithm, and improved region growth method. Systematic experiments demonstrate that our proposed segmentation network achieves superior accuracy compared to other competing methods. Additionally, the computed GBM thickness, after an automatic screening of suitable fields of view, meets the requirements for pathology evaluation.

GA-Net: A geographical attention neural network for the segmentation of body torso tissue composition

PurposeBody composition analysis (BCA) of the body torso plays a vital role in the study of physical health and pathology and provides biomarkers that facilitate the diagnosis and treatment of many diseases, such as type 2 diabetes mellitus, cardiovascular disease, obstructive sleep apnea, and osteoarthritis. In this work, we propose a body composition tissue segmentation method that can automatically delineate those key tissues, including subcutaneous adipose tissue, skeleton, skeletal muscle tissue, and visceral adipose tissue, on positron emission tomography/computed tomography scans of the body torso. MethodsTo provide appropriate and precise semantic and spatial information that is strongly related to body composition tissues for the deep neural network, first we introduce a new concept of the body area and integrate it into our proposed segmentation network called Geographical Attention Network (GA-Net). The body areas are defined following anatomical principles such that the whole body torso region is partitioned into three non-overlapping body areas. Each body composition tissue of interest is fully contained in exactly one specific minimal body area. Secondly, the proposed GA-Net has a novel dual-decoder schema that is composed of a tissue decoder and an area decoder. The tissue decoder segments the body composition tissues, while the area decoder segments the body areas as an auxiliary task. The features of body areas and body composition tissues are fused through a soft attention mechanism to gain geographical attention relevant to the body tissues. Thirdly, we propose a body composition tissue annotation approach that takes the body area labels as the region of interest, which significantly improves the reproducibility, precision, and efficiency of delineating body composition tissues. ResultsOur evaluations on 50 low-dose unenhanced CT images indicate that GA-Net outperforms other architectures statistically significantly based on the Dice metric. GA-Net also shows improvements for the 95% Hausdorff Distance metric in most comparisons. Notably, GA-Net exhibits more sensitivity to subtle boundary information and produces more reliable and robust predictions for such structures, which are the most challenging parts to manually mend in practice, with potentially significant time-savings in the post hoc correction of these subtle boundary placement errors. Due to the prior knowledge provided from body areas, GA-Net achieves competitive performance with less training data. Our extension of the dual-decoder schema to TransUNet and 3D U-Net demonstrates that the new schema significantly improves the performance of these classical neural networks as well. Heatmaps obtained from attention gate layers further illustrate the geographical guidance function of body areas for identifying body tissues. Conclusions(i) Prior anatomic knowledge supplied in the form of appropriately designed anatomic container objects significantly improves the segmentation of bodily tissues. (ii) Of particular note are the improvements achieved in the delineation of subtle boundary features which otherwise would take much effort for manual correction. (iii) The method can be easily extended to existing networks to improve their accuracy for this application.

A system-on-chip solution for deep learning-based automatic fetal biometric measurement

Ultrasound assessment of fetal biometry is widely used for gestational age prediction and abnormal fetal growth diagnosis. However, manual biometric measurements require time-consuming procedures and current automated measurement approaches remain challenging in terms of accuracy and computational complexity. This paper proposes deep learning-based efficient automatic fetal biometry measurement method for system-on-chip (SoC) solution. For end-to-end automated measurements, a hardware-friendly bilateral segmentation network (H-BiSeNet) was designed and optimized for multiple fetal objects (i.e., head, abdomen and femur), and fetal biometric parameters were then calculated with a robust multiple keypoint detection algorithm. For hardware implementation with a SoC, a deep learning processing unit (DPU) was employed to accelerate the proposed segmentation network, and biometric measurements with image pre- and postprocessing were conducted on an application processing unit (APU). A total of 6000 fetal ultrasound images in the three regions were collected from 2568 subjects, and the dataset was utilized to train the proposed network. In the performance evaluation, the proposed segmentation network with optimal quantization (H-BiSeNet-Q) outperformed other segmentation networks (i.e., U-Net, U-Net++, Attention U-Net, U-Net3+, DeepLabv3 + and BiSeNetv2) in terms of the average Dice coefficient, and automated measurements with H-BiSeNet-Q were highly correlated with manual measurements (r ≥ 0.983, p < 0.01 for all parameters, i.e., biparietal diameter, head circumference, abdominal circumference and femur length). In the execution time analysis, H-BiSeNet-Q ran at 4.25 frames per second (FPS) with parallelism corresponding to 80% resource utilization in the SoC. These results showed feasibility in low-resource hardware settings such as portable ultrasound systems.

A deep segmentation network for crack detection with progressive and hierarchical context fusion

Accurate detection of crack defects in infrastructure is crucial to ensure their safety and extend their service life. However, the presence of complex backgrounds, various shapes and sizes of crack defects, incomplete discontinuous crack defects, and class imbalance makes this task challenging. Traditional image processing techniques are sensitive to image noise and may miss cracks due to weak textures, complex lighting conditions, and other similar objects on the pavement. In recent years, deep learning-based segmentation networks have been proposed for crack defect detection, but they still face challenges in high-precision crack segmentation due to insufficient local feature processing, information loss caused by pooling operations, and limited receptive fields. To address these issues, we propose an end-to-end deep crack segmentation network, called PHCF-Net, which incorporates progressive and hierarchical context fusion. Firstly, the proposed network consists of progressive context fusion (PCF) and hierarchical context fusion (HCF) blocks for effective aggregation of global and local context information. Secondly, a multi-scale context fusion (MCF) block is proposed for multi-scale context extraction and aggregation. Finally, in order to solve the information loss problem caused by pooling operations, a hierarchical context fusion (HCF) block is proposed for effective aggregation of the deep and shallow features. In addition, a multi-scale input unit is also applied to the proposed segmentation network to obtain more context information. To evaluate the performance of PHCF-Net, we have conducted experiments on two publicly available crack segmentation datasets and compared its performance with mainstream segmentation models. The results demonstrate that proposed PHCF-Net achieves better pixel-level crack detection results and outperforms other advanced segmentation models.

Semi-supervised learning framework for crack segmentation based on contrastive learning and cross pseudo supervision

Fast and accurate crack segmentation plays an important role in the predictive maintenance of constructed facilities and civil infrastructures. However, it is worth noting that current deep-learning-based algorithms for crack segmentation may face significant challenges due to the requirement of a large amount of labeled data for high-precision segmentation. A novel semi-supervised learning framework for crack segmentation, which is referred to as semi-supervised crack (SemiCrack), based on the combination of contrastive learning and cross pseudo supervision (CPS) is presented in this study. The proposed segmentation network, called transformer and convolutional network (TC-Net), has a novel parallel encoder that fuses a transformer and a convolutional neural network. The inclusion of CPS can force the two models to maintain consistent outputs for various perturbed data based on the similarity loss. To capture the feature differences between positive and negative sample pairs extracted by the classifier and projector, pixel contrastive loss was also proposed. Compared with many state-of-the-art fully-supervised and semi-supervised segmentation algorithms, the results show that SemiCrack performs best on various publicly available datasets. The segmentation accuracy of TC-Net is higher than that of most fully-supervised segmentation networks, with an improvement of about 2% in Intersection of Union (IoU). Besides, SemiCrack requires only 20% labeled data to achieve comparable accuracy to other fully-supervised algorithms that require 100% labeled data. When the amount of labeled data is small, the IoUs of SemiCrack are significantly improved compared to fully supervised and semi-supervised networks.

An Attention-Guided CNN Framework for Segmentation and Grading of Glioma Using 3D MRI Scans.

Glioma has emerged as the deadliest form of brain tumor for human beings. Timely diagnosis of these tumors is a major step towards effective oncological treatment. Magnetic Resonance Imaging (MRI) typically offers a non-invasive inspection of brain lesions. However, manual inspection of tumors from MRI scans requires a large amount of time and it is also an error-prone process. Therefore, automated diagnosis of tumors plays a crucial role in clinical management and surgical interventions of gliomas. In this study, we propose a Convolutional Neural Network (CNN)-based framework for non-invasive grading of tumors from 3D MRI scans. The proposed framework incorporates two novel CNN architectures. The first CNN architecture performs the segmentation of tumors from multimodel MRI volumes. The proposed segmentation network leverages the spatial and channel attention modules to recalibrate the feature maps across the layers. The second network utilizes the multi-task learning strategy to perform the classification based on the three glioma grading tasks which include characterization of tumor into low-grade or high-grade, identification of 1p19q, and Isocitrate Dehydrogenase (IDH) status. We have carried out several experiments to evaluate the performance of our method. Extensive experimental observations indicate that the proposed framework achieves better performance than several state-of-the-art methods. We have also executed Welch's- t test to show the statistical significance of grading results. The source code of this study is available at https://github.com/prasunc/Gliomanet.

Predicting lung cancer treatment response from CT images using deep learning

AbstractLung cancer is the deadliest type of cancer and is one of the most frequently occurring cancers. It is primarily diagnosed in later stages when treatment becomes difficult. For better treatment and higher chances of survival, the treatment response of lung cancer patients needs to be analyzed to check whether the patients are responding to the treatment or not. This analysis can be done with the help of follow‐up computed tomography (CT) imaging before and after the treatment. However, manually analyzing the baseline and post‐treatment CT scan images of so many lung cancer patients is a tedious task. This study proposes an intuitive approach based on deep learning to analyze lung cancer through CT scan images before and after the treatment. In this approach, we utilized a segmentation network to segment the lung tumor in the follow‐up CT images. The segmented tumor is then analyzed to check the treatment effect, as suggested by the Response Evaluation Criteria in Solid Tumors (RECIST) guidelines. The segmentation network combines a vision transformer and a convolutional neural network. The segmentation network is first trained on a public dataset and then fine‐tuned on the local dataset to improve the segmentation performance. For this study, we have collected a lung cancer dataset from an Indian hospital. The dataset is divided into two parts dataset I and dataset II. Dataset I consists of 100 CT scans, which we use to fine‐tune the proposed segmentation network. Dataset II comprises 220 CT scans of 110 patients, consisting of baseline and post‐treatment scans. We use dataset II for testing. We achieved significant performance.

Multi-scale triple-attention network for pixelwise crack segmentation

Currently, intelligent crack detection is of great value for the maintenance of infrastructure, of which the most significant kind in China is roads. For pavement defects, the pavement can be repaired and maintained in a timely manner with an accurate defect detection task, which significantly reduces the occurrence of hazards. However, the detection of pavement defects remains a great challenge owing to many difficulties, for example, complex backgrounds, microdefects, various defect shapes and sizes, class imbalance issues, etc. Recently, deep learning has demonstrated its superior performance on pixelwise image segmentation, but some issues still exist on demanding pixelwise image segmentation, for instance, limited receptive field, insufficiency processing of local features, information loss issue generated by pooling operations, etc. Based on all of the above issues, a multiscale triple-attention network, named MST-Net, is proposed for end-to-end pixelwise crack detection. First, a multiscale input strategy is applied to the proposed segmentation network to capture more context information. Meanwhile, it can capably reduce the effect of the information loss issue generated by pooling operations. Second, to realize effective feature representation of local features, an additive attention fusion (AAF) block is proposed to guide feature learning to capture both global and local contexts. In addition, faced with the crack detection task with class imbalance issues, a triple attention (TA) block is proposed to detect spatial, channel and pixel attention information to suppress the background and useless information, which is conducive to the characterization of microcracks. Finally, aiming at the limited receptive field, a multiscale feature aggregation unit is proposed for feature fusion to increase the detection ability of multiscale defects. To better guide network training, a deep supervision mechanism is also introduced to speed up the convergence of the proposed segmentation model and improve the performance of defect segmentation. The related evaluation and detection experiments are carried out on three public datasets on crack segmentation, and the comparison experiments with the mainstream segmentation models show that the proposed segmentation network achieves excellent performance on pixelwise crack detection.

DEEP LEARNING-BASED HIGH-ACCURACY QUANTITATION FOR LUMBAR INTERVERTEBRAL DISC DEGENERATION FROM MRI

Using deep learning and image processing technology, a standardized automatic quantitative analysis systerm of lumbar disc degeneration based on T2MRI is proposed to help doctors evaluate the prognosis of intervertebral disc (IVD) degeneration.A semantic segmentation network BianqueNet with self-attention mechanism skip connection module and deep feature extraction module is proposed to achieve high-precision segmentation of intervertebral disc related areas. A quantitative method is proposed to calculate the signal intensity difference (SI) in IVD, average disc height (DH), disc height index (DHI), and disc height-to-diameter ratio (DHR). According to the correlation analysis results of the degeneration characteristic parameters of IVDs, 1051 MRI images from four hospitals were collected to establish the quantitative ranges for these IVD parameters in larger population around China.The average dice coefficients of the proposed segmentation network for vertebral bodies and intervertebral discs are 97.04% and 94.76%, respectively. The designed parameters of intervertebral disc degeneration have a significant negative correlation with the Modified Pfirrmann Grade. This procedure is suitable for different MRI centers and different resolution of lumbar spine T2MRI (ICC=.874~.958). Among them, the standard of intervertebral disc signal intensity degeneration has excellent reliability according to the modified Pfirrmann Grade (macroF1=90.63%~92.02%).we developed a fully automated deep learning-based lumbar spine segmentation network, which demonstrated strong versatility and high reliability to assist residents on IVD degeneration grading by means of IVD degeneration quantitation.

CNV-Net: Segmentation, Classification and Activity Score Measurement of Choroidal Neovascularization (CNV) Using Optical Coherence Tomography Angiography (OCTA)

This paper aims to present an artificial intelligence-based algorithm for the automated segmentation of Choroidal Neovascularization (CNV) areas and to identify the presence or absence of CNV activity criteria (branching, peripheral arcade, dark halo, shape, loop and anastomoses) in OCTA images. Methods: This retrospective and cross-sectional study includes 130 OCTA images from 101 patients with treatment-naïve CNV. At baseline, OCTA volumes of 6 × 6 mm2 were obtained to develop an AI-based algorithm to evaluate the CNV activity based on five activity criteria, including tiny branching vessels, anastomoses and loops, peripheral arcades, and perilesional hypointense halos. The proposed algorithm comprises two steps. The first block includes the pre-processing and segmentation of CNVs in OCTA images using a modified U-Net network. The second block consists of five binary classification networks, each implemented with various models from scratch, and using transfer learning from pre-trained networks. Results: The proposed segmentation network yielded an averaged Dice coefficient of 0.86. The individual classifiers corresponding to the five activity criteria (branch, peripheral arcade, dark halo, shape, loop, and anastomoses) showed accuracies of 0.84, 0.81, 0.86, 0.85, and 0.82, respectively. The AI-based algorithm potentially allows the reliable detection and segmentation of CNV from OCTA alone, without the need for imaging with contrast agents. The evaluation of the activity criteria in CNV lesions obtains acceptable results, and this algorithm could enable the objective, repeatable assessment of CNV features.

A high-performance deep-learning-based pipeline for whole-brain vasculature segmentation at the capillary resolution.

Reconstructing and analyzing all blood vessels throughout the brain is significant for understanding brain function, revealing the mechanisms of brain disease, and mapping the whole-brain vascular atlas. Vessel segmentation is a fundamental step in reconstruction and analysis. The whole-brain optical microscopic imaging method enables the acquisition of whole-brain vessel images at the capillary resolution. Due to the massive amount of data and the complex vascular features generated by high-resolution whole-brain imaging, achieving rapid and accurate segmentation of whole-brain vasculature becomes a challenge. We introduce HP-VSP, a high-performance vessel segmentation pipeline based on deep learning. The pipeline consists of three processes: data blocking, block prediction, and block fusion. We used parallel computing to parallelize this pipeline to improve the efficiency of whole-brain vessel segmentation. We also designed a lightweight deep neural network based on multi-resolution vessel feature extraction to segment vessels at different scales throughout the brain accurately. We validated our approach on whole-brain vascular data from three transgenic mice collected by HD-fMOST. The results show that our proposed segmentation network achieves the state-of-the-art level under various evaluation metrics. In contrast, the parameters of the network are only 1% of those of similar networks. The established segmentation pipeline could be used on various computing platforms and complete the whole-brain vessel segmentation in 3 h. We also demonstrated that our pipeline could be applied to the vascular analysis. The dataset is available at http://atlas.brainsmatics.org/a/li2301. The source code is freely available at https://github.com/visionlyx/HP-VSP.

Morphological prognosis prediction of choroid neovascularization from longitudinal SD-OCT images.

Choroid neovascularization (CNV) has no obvious symptoms in the early stage, but with its gradual expansion, leakage, rupture, and bleeding, it can cause vision loss and central scotoma. In some severe cases, it will lead to permanent visual impairment. Accurate prediction of disease progression can greatly help ophthalmologists to formulate appropriate treatment plans and prevent further deterioration of the disease. Therefore, we aim to predict the growth trend of CNV to help the attending physician judge the effectiveness of treatment. In this paper, we develop a CNN-based method for CNV growth prediction. To achieve this, we first design a registration network to rigidly register the spectral domain optical coherence tomography (SD-OCT) B-scans of each subject at different time points to eliminate retinal displacements of longitudinal data. Then, considering the correlation of longitudinal data, we propose a co-segmentation network with a correlation attention guidance (CAG) module to cooperatively segment CNV lesions of a group of follow-up images and use them as input for growth prediction. Finally, based on the above registration and segmentation networks, an encoder-recurrent-decoder framework is developed for CNV growth prediction, in which an attention-based gated recurrent unit (AGRU) is embedded as the recurrent neural network to recurrently learn robust representations. The registration network rigidly registers the follow-up images of patients to the reference images with a root mean square error (RMSE) of 6.754 pixels. And compared with other state-of-the-art segmentation methods, the proposed segmentation network achieves high performance with the Dice similarity coefficients (Dsc) of 85.27%. Based on the above experiments, the proposed growth prediction network can play a role in predicting the future CNV morphology, and the predicted CNV has a Dsc of 83.69% with the ground truth, which is significantly consistent with the actual follow-up visit. The proposed registration and segmentation networks provide the possibility for growth prediction. In addition, accurately predicting the growth of CNV enables us to know the efficacy of the drug against individuals in advance, creating opportunities for formulating appropriate treatment plans.

Where Shall I Touch? Vision-Guided Tactile Poking for Transparent Object Grasping

Picking up transparent objects is still a challenging task for robots. The visual properties of transparent objects such as reflection and refraction make the current grasping methods that rely on camera sensing fail to detect and localize them. However, humans can handle the transparent object well by first observing its coarse profile and then poking an area of interest to get a fine profile for grasping. Inspired by this, we propose a novel framework of vision-guided tactile poking for transparent objects grasping. In the proposed framework, a segmentation network is first used to predict the horizontal upper regions named as <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">poking regions</i> , where the robot can poke the object to obtain a good tactile reading, while leading to minimal disturbance to the object's state. A poke is then performed with a high-resolution GelSight tactile sensor. Given the local profiles improved with the tactile reading, a heuristic grasp is planned for grasping the transparent object. To mitigate the limitations of real-world data collection and labeling for transparent objects, a large-scale realistic synthetic dataset was constructed. Extensive experiments demonstrate that our proposed segmentation network can predict the potential poking region with a high mean average precision (mAP) of 0.360, and the vision-guided tactile poking can enhance the grasping success rate significantly from 38.9% to 85.2%. Thanks to its simplicity, our proposed approach could also be adopted by other force or tactile sensors and could be used for grasping of other challenging objects.

Cardiac LGE MRI Segmentation With Cross-Modality Image Augmentation and Improved U-Net.

Image segmentation is a challenging problem in imaging informatics, which stems from the intersection of imaging techniques, computer science and biomedicine. In particular, accurate segmentation of cardiac structures in late gadolinium enhancement (LGE) cardiac magnetic resonance (CMR) is of great clinical importance for cardiac function assessment and myocardial disease diagnosis. However, it is a well-known challenge due to its special imaging modality and the lack of labeled LGE samples. In this paper, we propose an unsupervised ventricular segmentation algorithm that can perform biventricular segmentation of LGE images in the absence of labeled LGE data. There are two primary modules, the data augmentation procedure and the segmentation network. The easily available annotated balanced-Steady State Free Precession (bSSFP) images are employed for cross-modal data augmentation by image translation, where a single bSSFP image is converted into multiple synthetic LGE images while preserving the original morphological structure. Then, the proposed segmentation network is trained with the synthetic LGE images and used for segmenting real LGE images. Validation experiments demonstrated the effectiveness and advantages of the proposed algorithm.

DGNet: Distribution Guided Efficient Learning for Oil Spill Image Segmentation

Successful implementation of oil spill segmentation in synthetic aperture radar (SAR) images is vital for marine environmental protection. In this article, we develop an effective segmentation framework named DGNet, which performs oil spill segmentation by incorporating the intrinsic distribution of backscatter values in SAR images. Specifically, our proposed segmentation network is constructed with two deep neural modules running in an interactive manner, where one is the inference module to achieve latent feature variable inference from SAR images and the other is the generative module to produce oil spill segmentation maps by drawing the latent feature variables as inputs. Thus, to yield accurate segmentation, we take into account the intrinsic distribution of backscatter values in SAR images and embed it in our segmentation model. The intrinsic distribution originates from SAR imagery, describing the physical characteristics of oil spills. In the training process, the formulated intrinsic distribution guides efficient learning of optimal latent feature variable inference for oil spill segmentation. The efficient learning enables the training of our proposed DGNet with a small amount of image data. This is economically beneficial to oil spill segmentation where the availability of oil spill SAR image data is limited in practice. Additionally, benefiting from optimal latent feature variable inference, our proposed DGNet performs accurate oil spill segmentation. We evaluate the segmentation performance of our proposed DGNet with different metrics, and experimental evaluations demonstrate its effective segmentations.

Deeply supervised vestibule segmentation network for CT images with global context‐aware pyramid feature extraction

Accurate vestibule segmentation for CT images is of great significance for the clinical diagnosis of congenital ear malformations and cochlear implant. However, it is still a challenging task due to extremely small size and irregular shape of vestibule. Here, a vestibule segmentation network for CT images is proposed under the basic encoder-decoder framework. Firstly, a residual block based on channel attention mechanism, named Res-CA block, is designed to guide the network to enhance the important features for the segmentation tasks while suppressing the irrelevant ones. And then, a global context-aware pyramid feature extraction (GCPFE) module is proposed to capture multi-receptive-field global context information. Finally, active contour with elastic (ACE) loss function is adopted to guide network learning more detailed information of the boundary. Furthermore, deep supervision (DS) mechanism is employed to locate the boundaries finely, improving the robustness of the network. The experiments are conducted on the self-established VestibuleDataset and UHRCT-Dataset, as well as publicly available retinal dataset, namely DRIVE, to comprehensively verify the robustness and generalization capability of the proposed segmentation network. The experimental results show that the proposed network can achieve a superior performance.

Cov-TransNet: Dual branch fusion network with transformer for COVID-19 infection segmentation

Segmentation of COVID-19 infection is a challenging task due to the blurred boundaries and low contrast between the infected and the non-infected areas in COVID-19 CT images, especially for small infection regions. COV-TransNet is presented to achieve high-precision segmentation of COVID-19 infection regions in this paper. The proposed segmentation network is composed of the auxiliary branch and the backbone branch. The auxiliary branch network adopts transformer to provide global information, helping the convolution layers in backbone branch to learn specific local features better. A multi-scale feature attention module is introduced to capture contextual information and adaptively enhance feature representations. Specially, a high internal resolution is maintained during the attention calculation process. Moreover, feature activation module can effectively reduce the loss of valid information during sampling. The proposed network can take full advantage of different depth and multi-scale features to achieve high sensitivity for identifying lesions of varied sizes and locations. We experiment on several datasets of the COVID-19 lesion segmentation task, including COVID-19-CT-Seg, UESTC-COVID-19, MosMedData and COVID-19-MedSeg. Comprehensive results demonstrate that COV-TransNet outperforms the existing state-of-the-art segmentation methods and achieves better segmentation performance for multi-scale lesions.

Deep learning-based computed tomography image segmentation and volume measurement of intracerebral hemorrhage.

The study aims to enhance the accuracy and practicability of CT image segmentation and volume measurement of ICH by using deep learning technology. A dataset including the brain CT images and clinical data of 1,027 patients with spontaneous ICHs treated from January 2010 to December 2020 were retrospectively analyzed, and a deep segmentation network (AttFocusNet) integrating the focus structure and the attention gate (AG) mechanism is proposed to enable automatic, accurate CT image segmentation and volume measurement of ICHs. In internal validation set, experimental results showed that AttFocusNet achieved a Dice coefficient of 0.908, an intersection-over-union (IoU) of 0.874, a sensitivity of 0.913, a positive predictive value (PPV) of 0.957, and a 95% Hausdorff distance (HD95) (mm) of 5.960. The intraclass correlation coefficient (ICC) of the ICH volume measurement between AttFocusNet and the ground truth was 0.997. The average time of per case achieved by AttFocusNet, Coniglobus formula and manual segmentation is 5.6, 47.7, and 170.1 s. In the two external validation sets, AttFocusNet achieved a Dice coefficient of 0.889 and 0.911, respectively, an IoU of 0.800 and 0.836, respectively, a sensitivity of 0.817 and 0.849, respectively, a PPV of 0.976 and 0.981, respectively, and a HD95 of 5.331 and 4.220, respectively. The ICC of the ICH volume measurement between AttFocusNet and the ground truth were 0.939 and 0.956, respectively. The proposed segmentation network AttFocusNet significantly outperforms the Coniglobus formula in terms of ICH segmentation and volume measurement by acquiring measurement results closer to the true ICH volume and significantly reducing the clinical workload.