Decoding Path Research Articles

Decoding Reed-Muller Codes With Successive Codeword Permutations

A novel recursive list decoding (RLD) algorithm for Reed-Muller (RM) codes based on successive permutations (SP) of the codeword is presented. A low-complexity SP scheme applied to a subset of the symmetry group of RM codes is first proposed to carefully select a good codeword permutation on the fly. Then, the proposed SP technique is integrated into an improved RLD algorithm that initializes different decoding paths with random codeword permutations, which are sampled from the full symmetry group of RM codes. Finally, efficient latency and complexity reduction schemes are introduced that virtually preserve the error-correction performance of the proposed decoder. Simulation results demonstrate that at the target frame error rate of 10−3 for the RM code of length 256 with 163 information bits, the proposed decoder reduces 6% of the computational complexity and 22% of the decoding latency of the state-of-the-art semi-parallel simplified successive-cancellation decoder with fast Hadamard transform (SSC-FHT) that uses 96 permutations from the full symmetry group of RM codes, while relatively maintaining the error-correction performance and memory consumption of the semi-parallel permuted SSC-FHT decoder.

A Change Detection Method Based on Multi-Scale Adaptive Convolution Kernel Network and Multimodal Conditional Random Field for Multi-Temporal Multispectral Images

Multispectral image change detection is an important application in the field of remote sensing. Multispectral images usually contain many complex scenes, such as ground objects with diverse scales and proportions, so the change detection task expects the feature extractor is superior in adaptive multi-scale feature learning. To address the above-mentioned problems, a multispectral image change detection method based on multi-scale adaptive kernel network and multimodal conditional random field (MSAK-Net-MCRF) is proposed. The multi-scale adaptive kernel network (MSAK-Net) extends the encoding path of the U-Net, and designs a weight-sharing bilateral encoding path, which simultaneously extracts independent features of bi-temporal multispectral images without introducing additional parameters. A selective convolution kernel block (SCKB) that can adaptively assign weights is designed and embedded in the encoding path of MSAK-Net to extract multi-scale features in images. MSAK-Net retains the skip connections in the U-Net, and embeds an upsampling module (UM) based on the attention mechanism in the decoding path, which can give the feature map a better expression of change information in both the channel dimension and the spatial dimension. Finally, the multimodal conditional random field (MCRF) is used to smooth the detection results of the MSAK-Net. Experimental results on two public multispectral datasets indicate the effectiveness and robustness of the proposed method when compared with other state-of-the-art methods.

FCSNet: A quantitative explanation method for surface scratch defects during belt grinding based on deep learning

The presence of surface defects during abrasive belt grinding has a significant impact on the service performance of the parts, and the precise extraction of surface defect feature information helps to enhance grinding quality. However, in the face of abrasive belt grinding surfaces with complex texture features, it is a challenging task to effectively extract defect features and quantify the characterization. Therefore, this research suggests a quantitative explanation approach for surface scratch defects in abrasive belt grinding based on the deep learning theory. Facing challenging issues like complex texture background, defect size, area, and depth of grinding surface, the automatic segmentation model (FCSNet) concentrating on the multiscale channel and spatial attention information is proposed. To concentrate on the multiscale channel and spatial information and accurately detect the defect features, the residual atrous convolutional pyramid module with channel and spatial dual attention module (RAPCS) is constructed. To capture the global, long-range context characteristics and emphasize target regions, the convolutional block attention module (CBAM) concatenate block is included in the skip connections between the encoder path and the decoder path. To address the issue of foreground-background data imbalance caused by too small segmented target regions, the Focal Tversky hybrid loss function was adopted to guide the model to learn the difficult region of interest (ROI). At the end of the network structure, Grad-CAM is adopted and statistical knowledge is incorporated for a qualitative and quantitative visual explanation of the segmentation results. Finally, the dataset of surface scratch defects in abrasive belt grinding was established and numerous experiments were carried out based on this dataset. The results demonstrate that the proposed model has an excellent segmentation performance and defect quantitative explanation ability, achieving 98.80% in Accuracy, 81.72% in Recall, 81.81% in Precision, 81.42% in F1-score, and 81.87% in mIoU, respectively.

N-Net: Lesion region segmentations using the generalized hybrid dilated convolutions for polyps in colonoscopy images.

Colorectal cancer is the cancer with the second highest and the third highest incidence rates for the female and the male, respectively. Colorectal polyps are potential prognostic indicators of colorectal cancer, and colonoscopy is the gold standard for the biopsy and the removal of colorectal polyps. In this scenario, one of the main concerns is to ensure the accuracy of lesion region identifications. However, the missing rate of polyps through manual observations in colonoscopy can reach 14%–30%. In this paper, we focus on the identifications of polyps in clinical colonoscopy images and propose a new N-shaped deep neural network (N-Net) structure to conduct the lesion region segmentations. The encoder-decoder framework is adopted in the N-Net structure and the DenseNet modules are implemented in the encoding path of the network. Moreover, we innovatively propose the strategy to design the generalized hybrid dilated convolution (GHDC), which enables flexible dilated rates and convolutional kernel sizes, to facilitate the transmission of the multi-scale information with the respective fields expanded. Based on the strategy of GHDC designing, we design four GHDC blocks to connect the encoding and the decoding paths. Through the experiments on two publicly available datasets on polyp segmentations of colonoscopy images: the Kvasir-SEG dataset and the CVC-ClinicDB dataset, the rationality and superiority of the proposed GHDC blocks and the proposed N-Net are verified. Through the comparative studies with the state-of-the-art methods, such as TransU-Net, DeepLabV3+ and CA-Net, we show that even with a small amount of network parameters, the N-Net outperforms with the Dice of 94.45%, the average symmetric surface distance (ASSD) of 0.38 pix and the mean intersection-over-union (mIoU) of 89.80% on the Kvasir-SEG dataset, and with the Dice of 97.03%, the ASSD of 0.16 pix and the mIoU of 94.35% on the CVC-ClinicDB dataset.

Global and Local Feature Reconstruction for Medical Image Segmentation.

Learning how to capture long-range dependencies and restore spatial information of down-sampled feature maps are the basis of the encoder-decoder structure networks in medical image segmentation. U-Net based methods use feature fusion to alleviate these two problems, but the global feature extraction ability and spatial information recovery ability of U-Net are still insufficient. In this paper, we propose a Global Feature Reconstruction (GFR) module to efficiently capture global context features and a Local Feature Reconstruction (LFR) module to dynamically up-sample features, respectively. For the GFR module, we first extract the global features with category representation from the feature map, then use the different level global features to reconstruct features at each location. The GFR module establishes a connection for each pair of feature elements in the entire space from a global perspective and transfers semantic information from the deep layers to the shallow layers. For the LFR module, we use low-level feature maps to guide the up-sampling process of high-level feature maps. Specifically, we use local neighborhoods to reconstruct features to achieve the transfer of spatial information. Based on the encoder-decoder architecture, we propose a Global and Local Feature Reconstruction Network (GLFRNet), in which the GFR modules are applied as skip connections and the LFR modules constitute the decoder path. The proposed GLFRNet is applied to four different medical image segmentation tasks and achieves state-of-the-art performance.

Res-CDD-Net: A Network with Multi-Scale Attention and Optimized Decoding Path for Skin Lesion Segmentation

Melanoma is a lethal skin cancer. In its diagnosis, skin lesion segmentation plays a critical role. However, skin lesions exhibit a wide range of sizes, shapes, colors, and edges. This makes skin lesion segmentation a challenging task. In this paper, we propose an encoding–decoding network called Res-CDD-Net to address the aforementioned aspects related to skin lesion segmentation. First, we adopt ResNeXt50 pre-trained on the ImageNet dataset as the encoding path. This pre-trained ResNeXt50 can provide rich image features to the whole network to achieve higher segmentation accuracy. Second, a channel and spatial attention block (CSAB), which integrates both channel and spatial attention, and a multi-scale capture block (MSCB) are introduced between the encoding and decoding paths. The CSAB can highlight the lesion area and inhibit irrelevant objects. MSCB can extract multi-scale information to learn lesion areas of different sizes. Third, we upgrade the decoding path. Every 3 × 3 square convolution kernel in the decoding path is replaced by a diverse branch block (DBB), which not only promotes the feature restoration capability, but also improves the performance and robustness of the network. We evaluate the proposed network on three public skin lesion datasets, namely ISIC-2017, ISIC-2016, and PH2. The dice coefficient is 6.90% higher than that of U-Net, whereas the Jaccard index is 10.84% higher than that of U-Net (assessed on the ISIC-2017 dataset). The results show that Res-CDD-Net achieves outstanding performance, higher than the performance of most state-of-the-art networks. Last but not least, the training of the network is fast, and good results can be achieved in early stages of training.

MM-UNet: A multimodality brain tumor segmentation network in MRI images

The global annual incidence of brain tumors is approximately seven out of 100,000, accounting for 2% of all tumors. The mortality rate ranks first among children under 12 and 10th among adults. Therefore, the localization and segmentation of brain tumor images constitute an active field of medical research. The traditional manual segmentation method is time-consuming, laborious, and subjective. In addition, the information provided by a single-image modality is often limited and cannot meet the needs of clinical application. Therefore, in this study, we developed a multimodality feature fusion network, MM-UNet, for brain tumor segmentation by adopting a multi-encoder and single-decoder structure. In the proposed network, each encoder independently extracts low-level features from the corresponding imaging modality, and the hybrid attention block strengthens the features. After fusion with the high-level semantic of the decoder path through skip connection, the decoder restores the pixel-level segmentation results. We evaluated the performance of the proposed model on the BraTS 2020 dataset. MM-UNet achieved the mean Dice score of 79.2% and mean Hausdorff distance of 8.466, which is a consistent performance improvement over the U-Net, Attention U-Net, and ResUNet baseline models and demonstrates the effectiveness of the proposed model.

Color-invariant skin lesion semantic segmentation based on modified U-Net deep convolutional neural network.

Melanoma is a type of skin lesion that is less common than other types of skin lesions, but it is fast growing and spreading. Therefore, it is classified as a serious disease that directly threatens human health and life. Recently, the number of deaths due to this disease has increased significantly. Thus, researchers are interested in creating computer-aided diagnostic systems that aid in the proper diagnosis and detection of these lesions from dermoscopy images. Relying on manual diagnosis is time consuming in addition to requiring enough experience from dermatologists. Current skin lesion segmentation systems use deep convolutional neural networks to detect skin lesions from RGB dermoscopy images. However, relying on RGB color model is not always the optimal choice to train such networks because some fine details of lesion parts in the dermoscopy images can not clearly appear using RGB color model. Other color models exhibit invariant features of the dermoscopy images so that they can improve the performance of deep neural networks. In the proposed Color Invariant U-Net (CIU-Net) model, a color mixture block is added at the beginning of the contracting path of U-Net. The color mixture block acts as a mixer to learn the fusion of various input color models and create a new one with three channels. Furthermore, a new channel-attention module is included in the connection path between encoder and decoder paths. This channel attention module is developed to enrich the extracted color features. From the experimental result, we found that the proposed CIU-Net works in harmony with the new proposed hybrid loss function to enhance skin segmentation results. The performance of the proposed CIU-Net architecture is evaluated using ISIC 2018 dataset and the results are compared with other recent approaches. Our proposed method outperformed other recent approaches and achieved the best Dice and Jaccard coefficient with values 92.56% and 91.40%, respectively.

Improved Belief Propagation List Decoding for Polar Codes

Polar codes have become the channel coding scheme for control channel of enhanced mobile broadband in the 5G communication systems. Belief propagation (BP) decoding of polar codes has advantages of low decoding latency and high parallelism but achieves worse bit error ratio (BER) performance compared with the successive cancellation list (SCL) decoding scheme. In this paper, an improved BP list (IBPL) decoding algorithm is proposed with comparable BER performance to SCL algoritm. Firstly, the optimal permuted factor graph is analyzed for polar codes, which improves the performance of the BP decoder without path extension. Furthermore, based on the optimal graph, the bit metric and decoding path metric are proposed to extend and prune the decoding path. The proposed IBPL decoder is focused on not only the permutation of polar codes but also the reliabilities of decoded codewords during each iteration of BP decoding, which has a more accurate decoding path list. The simulation results show that the proposed IBPL decoder improves the BER performance compared with the original BP decoder significantly, and can approach the performance of the SCL decoder at low signal to noise ratio regions.

Joint Successive Cancellation List Decoding for the Double Polar Codes

As a new joint source-channel coding scheme, the double polar (D-Polar) codes have been proposed recently. In this letter, a novel joint source-channel decoder, namely the joint successive cancellation list (J-SCL) decoder, is proposed to improve the decoding performance of the D-Polar codes. We merge the trellis of the source polar code and that of the channel polar code to construct a compound trellis. In this compound trellis, the variable nodes corresponding to the information bits of the channel polar code and the variable nodes representing the high-entropy bits are merged into the joint source-channel (JSC) nodes. Based on the compound trellis, the J-SCL decoder is designed to recover the source messages by combining the source SCL decoding and channel SCL decoding. The proposed J-SCL decoder doubles the number of the decoding paths for each JSC node and low-entropy node, and then discard all but the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$L$ </tex-math></inline-formula> paths with the smallest joint path-metric (JPM). For the JSC node, the JPM is updated considering both the channel decision log-likelihood ratios (LLRs) and the source decision LLRs. Simulation results show that the J-SCL decoder outperforms the turbo-like BP (TL-BP) decoder with lower complexity.

Stacked dilated convolutions and asymmetric architecture for U-Net-based medical image segmentation

Deep learning has been widely utilized for medical image segmentation. The most commonly used U-Net and its variants often share two common characteristics but lack solid evidence for the effectiveness. First, each block (i.e., consecutive convolutions of feature maps of the same resolution) outputs feature maps from the last convolution, limiting the variety of the receptive fields. Second, the network has a symmetric structure where the encoder and the decoder paths have similar numbers of channels. We explored two novel revisions: a stacked dilated operation that outputs feature maps from multi-scale receptive fields to replace the consecutive convolutions; an asymmetric architecture with fewer channels in the decoder path. Two novel models were developed: U-Net using the stacked dilated operation (SDU-Net) and asymmetric SDU-Net (ASDU-Net). We used both publicly available and private datasets to assess the efficacy of the proposed models. Extensive experiments confirmed SDU-Net outperformed or achieved performance similar to the state-of-the-art while using fewer parameters (40% of U-Net). ASDU-Net further reduced the model parameters to 20% of U-Net with performance comparable to SDU-Net. In conclusion, the stacked dilated operation and the asymmetric structure are promising for improving the performance of U-Net and its variants.

W-Net: Novel Deep Supervision for Deep Learning-based Cardiac Magnetic Resonance Imaging Segmentation

Cardiac magnetic resonance imaging (CMRI) segmentation transforms cardiac MR images into semantic regions to define the left ventricle cavity, right ventricle cavity, and myocardium. CMRI segmentation provides ventricles’ volume, mass, and ejection fraction, playing a significant role in cardiac disease diagnosis. This paper aims to propose novel deep supervision in U-Net-based architecture to enhance the segmentation performance. It presents a deeply supervised W-Net, which creates another path in parallel with the decoder path in U-Net-based architecture. The output of every upsampling layer in the decoder path is combined with pixel-wise addition for feature reuse, and loss is computed at each feature dimension on the deep supervision layer, which enables gradients to be implanted at a greater depth into the network and enhances the training of all layers in the network. Proposed W-Net applied on single scanner-based ACDC dataset and Multi-Centre, Multi-Vendor & Multi-Disease dataset, making it more robust in model generalization. W-Net significantly outperforms numerous state-of-the-art methods on the two publicly available CMRI datasets, according to experiments conducted. W-Net obtained better segmentation results ranked in the top three for many metrics. It is evident that the proposed W-Net has considerable potential in CMRI segmentation, cardiac assessment, and disease diagnosis.

CEL-Unet: a novel CNN architecture for 3D Segmentation of Knee Bones affected by Severe Osteoarthritis for PSI-Based Surgical Planning.

Unet architectures are promising deep learning networks exploited to perform the automatic segmentation of bone CT images, in line with their ability to deal with pathological deformations and size-varying anatomies. However, bone degeneration, like the development of irregular osteophytes as well as mineral density alterations might interfere with this automated process and demand extensive manual refinement. The aim of this work is to implement an innovative Unet variant, the CEL-Unet, to improve the femur and tibia segmentation outcomes in osteoarthritic knee joints. In this network the decoding path is split into a region and contour-aware branch to increase the prediction reliability in such pathological conditions. The comparison between the segmentation results achieved with a standard Unet and its novel variant (CEL-Unet) was performed as follows: the Unet was trained with 5 different loss functions: Dice Loss, Focal Loss, Exponential Logarithmic Loss, Double Cross Entropy Loss and Distanced Cross Entropy loss. The CEL-Unet was instead trained with two loss functions, one for each of the network outputs, namely Mask and Edge, yielding the so-called Combined Edge Loss (CEL) function. A set of 259 knee CT scans was used to train the model and test segmentation performance. The CEL-Unet outperformed all other Unet-based models, reaching the highest Jaccard values of about 0.97 and 0.96 on femur and tibia, respectively. Clinical Relevance- With the increasing rate of Total Knee Arthoplasty deep learning-based methods can achieve fast accurate and automatic 3D segmentation of the knee joint bones to enhance new costumized pre-operative planning.

An Improved U-Net Segmentation Model That Integrates a Dual Attention Mechanism and a Residual Network for Transformer Oil Leakage Detection

Accurately detecting oil leakage from a power transformer is important to maintain its normal operation. Deep learning (DL) methods have achieved satisfactory performance in automatic oil detection, but challenges remain due to the small amount of training data and oil targets with large variations in position, shape, and scale. To manage these issues, we propose a dual attention residual U-net (DAttRes-Unet) within a U-net architecture that extensively uses a residual network as well as spatial and channel-wise attention modules. To overcome the vanishing gradient problem due to deeper layers and a small amount of training data, a residual module from ResNet18 is used to construct the encoder path in the U-net framework. Meanwhile, to overcome the issue of training difficulty for the network, inspired by the advantage of transfer learning, initial network parameters in the encoder are obtained from the pre-trained ResNet18 on the ImageNet dataset. Further, in the decoder path, spatial attention and channel attention are integrated to highlight oil-stained regions while suppressing the background or irrelevant parts/channels. To facilitate the acquisition of the fluorescence images of the transformer, we designed a portable acquisition device integrating an ultraviolet light source and a digital camera. The proposed network is trained on the amount of fluorescence images after data augmentation is used and tested on actual fluorescence images. The experiment results show that the proposed DAttRes-Unet network can recognize oil-stained regions with a high accuracy of 98.49% for various shapes and scales of oil leakage.

DGCU–Net: A new dual gradient-color deep convolutional neural network for efficient skin lesion segmentation

Computer aided diagnosis (CAD) systems for Melanoma detection require an efficient skin lesion segmentation from dermoscopy images. However, irregular lesion shape and the large variations of colors have a significant impact on segmentation results. Therefore, methods developed for segmentation process have to be concerned with extracting boundaries and texture details of the lesion images with high accuracy. Recent deep learning convolutional neural networks such as U-Net achieves a significant progress in skin lesion segmentation. However, the segmentation results of original U-Net are not satisfactory to precisely find the borders of skin lesion. This paper presents a new Dual Gradient-Color U-Net (DGCU–Net) model which contains two encoders and a single decoder paths. The proposed DGCU–Net model integrates features extracted from an invariant color image representation with the gradient information of the input image to strengthen the borders of skin lesion. Gradient information of the input image is employed to calculate the attention maps used to strengthen color features in the first color encoder path. In addition, Atrous spatial pyramid pooling (ASPP) block is utilized in the bridge stage between encoder and decoder sub-networks. Spatial and channel attention network is also utilized to strengthen the connections between encoder and decoder blocks. The proposed network architecture significantly improves skin lesion segmentation results of the baseline U-Net model. Proposed model is evaluated using three benchmark datasets: ISIC 2017, ISIC 2018, and PH2 datasets. The comparison with state-of-the-art methods verified the superiority of our proposed DGCU–Net in various evaluation metrics.

U-Net applied to retrieve two-dimensional temperature and CO[formula omitted] concentration fields of laminar diffusion flames

Direct absorption spectroscopy measurement of the 4.3μm CO2 using interband cascade laser with a high spectral resolution, provides theoretical feasibility to retrieve the spatial distribution of CO2 temperature and concentration in flames. However, retrieving temperature and concentrations from the line-of-sight spectral measurements involves solving non-linear and ill-posed problems. Traditional reconstruction algorithms for axisymmetric flames require multiple line-of-sight measurements in both axial and radial directions. The complicated experimental system limits the reconstruction efficiency. Here, a novel reconstruction model is proposed and demonstrated based on the U-Net model, a fully convolutional network model composed by a encoder–decoder path forming a symmetric U-shaped architecture, which only requires measurements of the central axis of the flame to simultaneously and efficiently reconstruct the two-dimensional temperature and CO2 concentration fields for axisymmetric flames. The U-Net model is trained using datasets from simulated temperature, CO2 concentration and their corresponding spectral optical thickness. The proposed U-Net model has been applied to reconstruct the two-dimensional temperature and CO2 concentration fields of two previously measured laminar diffusion flames. Comparing with results from the traditional Abel inversion and two-line atomic fluorescence thermometry methods, the U-Net reconstruction model can better preserve the spatial continuity and correlation of flame temperature and CO2 concentration from information hidden in the flame spectral images, which considerably simplifies the measurement system and improves the reconstruction performance.

Automatic microscopic diagnosis of diseases using an improved UNet++ architecture

Anthrax is a severe infectious disease caused by the Bacillus anthracis bacterium. This paper aims to design and implement a fast and reliable system based on microscopic image processing of patient tissue samples for the automatic diagnosis of anthrax and other tissues diseases, metastasis detection, patient prognosis, etc. An improved UNet++ architecture is proposed to segment microscopic images of patient tissue samples. The proposed model combines multi-scale features by adding skip connections in two paths; the forward path from the encoder to the decoder and the decoder path to the output. These new connections improve the performance of the UNet++. Integration of the squeeze and excitation-inception blocks in the new skip connections provides the network with features at different scales with different kernel sizes. Several convolutional networks are used as the backbone to extract powerful representations in the encoder section. The use of batch normalization, dropout technique, and LRelu activation function in this model accelerates convergence and increases the generalization power of the model. To overcome the problem of data imbalance of different classes, a weighted hybrid loss function is proposed, which further improved segmentation efficiency. The semantic segmentation results are converted to the instance segmentation using the marker-based watershed algorithm. Experimental results show that despite many challenges of microscopic image analysis, the proposed model is a reliable system for the automatic diagnosis of anthrax and other tissues diseases. It produces better results than state-of-the-art architectures.

Deep U-Net architecture with curriculum learning for myocardial pathology segmentation in multi-sequence cardiac magnetic resonance images

Myocardial pathology segmentation is essential for the diagnosis and treatment of patients suffering from myocardial infarction. In this work, we propose an end-to-end deep learning based segmentation method for automatically delineating the area of left ventricle (LV) myocardial infarct and edema regions. The proposed method uses the 6 layers deep U-Net architecture as the segmentation backbone, which adopts a hierarchical feature representation with symmetrical encoder–decoder paths. Skip connections are added between encoder and decoder paths, to concatenate low-level and high-level information for better feature representation. Moreover, three other modules, direction field module (DFM), channel self-attention module (CAM) and selective kernel module (SKM), have also been implemented for further exploration of performance improvement. The proposed method is tested on the public MyoPS 2020 (myocardial pathology segmentation combining multi-sequence cardiac magnetic resonance) challenge dataset. Compared with extra self-attention module or selective kernel module, plain deep U-Net with curriculum learning achieves better results on testing dataset. Extensive ablation experiments are performed to explore the optimal depth of U-Net, multiple loss functions and different data augmentation methods. Using the official evaluation kit, our solution outperforms state-of-the-art single stage approaches, and achieves comparable performance with other advanced multi-stage methods. The evaluation results demonstrate our method’s effectiveness on myocardial pathology segmentation in multi-sequence cardiac magnetic resonance (CMR) data, and the superiority to the current state-of-the-art single stage methods.

Attention-modulated multi-branch convolutional neural networks for neonatal brain tissue segmentation

Accurate measurement of brain structures is essential for the evaluation of neonatal brain growth and development. The conventional methods use manual segmentation to measure brain tissues, which is very time-consuming and inefficient. Recent deep learning achieves excellent performance in computer vision, but it is still unsatisfactory for segmenting magnetic resonance images of neonatal brains because they are immature with unique attributes. In this paper, we propose a novel attention-modulated multi-branch convolutional neural network for neonatal brain tissue segmentation. The proposed network is built on the encoder-decoder framework by introducing both multi-scale convolutions in the encoding path and multi-branch attention modules in the decoding path. Multi-scale convolutions with different kernels are used to extract rich semantic features across large receptive fields in the encoding path. Multi-branch attention modules are used to capture abundant contextual information in the decoding path for segmenting brain tissues by fusing both local features and their corresponding global dependencies. Spatial attention connections between the encoding and decoding paths are designed to increase feature propagation for both avoiding information loss during downsampling and accelerating model training convergence. The proposed network was implemented in comparison with baseline methods on three neonatal brain datasets. Our network achieves the average Dice similarity coefficients/the average Hausdorff distances of 0.9116/8.1289, 0.9367/9.8212 and 0.8931/8.1612 on the customized dCBP2021 dataset, 0.8786/11.7863, 0.8965/13.4296 and 0.8539/10.462 on the public NBAtlas dataset, as well as 0.9253/7.7968, 0.9448/9.5472 and 0.9132/7.5877 on the public dHCP2017 dataset in partitioning the brain into gray matter, white matter and cerebrospinal fluid, respectively. The experimental results show that the proposed method achieves competitive state-of-the-art performance in neonatal brain tissue segmentation. The code and pre-trained models are available at https://github.com/zhangyongqin/AMCNN.

CEL-Unet: Distance Weighted Maps and Multi-Scale Pyramidal Edge Extraction for Accurate Osteoarthritic Bone Segmentation in CT Scans

Unet architectures are being investigated for automatic image segmentation of bones in CT scans because of their ability to address size-varying anatomies and pathological deformations. Nonetheless, changes in mineral density, narrowing of joint spaces and formation of largely irregular osteophytes may easily disrupt automatism requiring extensive manual refinement. A novel Unet variant, called CEL-Unet, is presented to boost the segmentation quality of the femur and tibia in the osteoarthritic knee joint. The neural network embeds region-aware and two contour-aware branches in the decoding path. The paper features three main technical novelties: 1) directed connections between contour and region branches progressively at different decoding scales; 2) pyramidal edge extraction in the contour branch to perform multi-resolution edge processing; 3) distance-weighted cross-entropy loss function to increase delineation quality at the sharp edges of the shapes. A set of 700 knee CT scans was used to train the model and test segmentation performance. Qualitatively CEL-Unet correctly segmented cases where the state-of-the-art architectures failed. Quantitatively, the Jaccard indexes of femur and tibia segmentation were 0.98 and 0.97, with median 3D reconstruction errors less than 0.80 and 0.60 mm, overcoming competitive Unet models. The results were evaluated against knee arthroplasty planning based on personalized surgical instruments (PSI). Excellent agreement with reference data was found for femoral (0.11°) and tibial (0.05°) alignments of the distal and proximal cuts computed on the reconstructed surfaces. The bone segmentation was effective for large pathological deformations and osteophytes, making the techniques potentially usable in PSI-based surgical planning, where the reconstruction accuracy of the bony shapes is one of the main critical factors for the success of the operation.