Light3DHS: A lightweight 3D hippocampus segmentation method using multiscale convolution attention and vision transformer.

The morphological analysis and volume measurement of the hippocampus are crucial to the study of many brain diseases. Therefore, an accurate hippocampal segmentation method is beneficial for the development of clinical research in brain diseases. U-Net and its variants have become prevalent in hippocampus segmentation of Magnetic Resonance Imaging (MRI) due to their effectiveness, and the architecture based on Transformer has also received some attention. However, some existing methods focus too much on the shape and volume of the hippocampus rather than its spatial information, and the extracted information is independent of each other, ignoring the correlation between local and global features. In addition, many methods cannot be effectively applied to practical medical image segmentation due to many parameters and high computational complexity. To this end, we combined the advantages of CNNs and ViTs (Vision Transformer) and proposed a simple and lightweight model: Light3DHS for the segmentation of the 3D hippocampus. In order to obtain richer local contextual features, the encoder first utilizes a multi-scale convolutional attention module (MCA) to learn the spatial information of the hippocampus. Considering the importance of local features and global semantics for 3D segmentation, we used a lightweight ViT to learn high-level features of scale invariance and further fuse local-to-global representation. To evaluate the effectiveness of encoder feature representation, we designed three decoders of different complexity to generate segmentation maps. Experiments on three common hippocampal datasets demonstrate that the network achieves more accurate hippocampus segmentation with fewer parameters. Light3DHS performs better than other state-of-the-art algorithms.

A Lightweight Convolutional Neural Network Based on Dynamic Level-Set Loss Function for Spine MR Image Segmentation.

BACKGROUND: Spine MR image segmentation is important foundation for computer-aided diagnostic (CAD) algorithms of spine disorders. Convolutional neural networks segment effectively, but require high computational costs. PURPOSE: To design a lightweight model based on dynamic level-set loss function for high segmentation performance. STUDY TYPE: Retrospective. POPULATION: Four hundred forty-eight subjects (3163 images) from two separate datasets. Dataset-1: 276 subjects/994 images (53.26% female, mean age 49.02 ± 14.09), all for disc degeneration screening, 188 had disc degeneration, 67 had herniated disc. Dataset-2: public dataset with 172 subjects/2169 images, 142 patients with vertebral degeneration, 163 patients with disc degeneration. FIELD STRENGTH/SEQUENCE: T2 weighted turbo spin echo sequences at 3T. ASSESSMENT: Dynamic Level-set Net (DLS-Net) was compared with four mainstream (including U-net++) and four lightweight models, and manual label made by five radiologists (vertebrae, discs, spinal fluid) used as segmentation evaluation standard. Five-fold cross-validation are used for all experiments. Based on segmentation, a CAD algorithm of lumbar disc was designed for assessing DLS-Net's practicality, and the text annotation (normal, bulging, or herniated) from medical history data were used as evaluation standard. STATISTICAL TESTS: All segmentation models were evaluated with DSC, accuracy, precision, and AUC. The pixel numbers of segmented results were compared with manual label using paired t-tests, with P < 0.05 indicating significance. The CAD algorithm was evaluated with accuracy of lumbar disc diagnosis. RESULTS: With only 1.48% parameters of U-net++, DLS-Net achieved similar accuracy in both datasets (Dataset-1: DSC 0.88 vs. 0.89, AUC 0.94 vs. 0.94; Dataset-2: DSC 0.86 vs. 0.86, AUC 0.93 vs. 0.93). The segmentation results of DLS-Net showed no significant differences with manual labels in pixel numbers for discs (Dataset-1: 1603.30 vs. 1588.77, P = 0.22; Dataset-2: 863.61 vs. 886.4, P = 0.14) and vertebrae (Dataset-1: 3984.28 vs. 3961.94, P = 0.38; Dataset-2: 4806.91 vs. 4732.85, P = 0.21). Based on DLS-Net's segmentation results, the CAD algorithm achieved higher accuracy than using non-cropped MR images (87.47% vs. 61.82%). DATA CONCLUSION: The proposed DLS-Net has fewer parameters but achieves similar accuracy to U-net++, helps CAD algorithm achieve higher accuracy, which facilitates wider application. EVIDENCE LEVEL: 2 TECHNICAL EFFICACY: Stage 1.

Improving domain generalization performance for medical image segmentation via random feature augmentation.

Deep convolutional neural networks (DCNNs) have shown remarkable performance in medical image segmentation tasks. However, medical images frequently exhibit distribution discrepancies due to variations in scanner vendors, operators, and image quality, which pose significant challenges to the robustness of trained models when applied to unseen clinical data. To address this issue, domain generalization methods have been developed to enhance the generalization ability of DCNNs. Feature space-based data augmentation methods have been proven effective in improving domain generalization, but they often rely on prior knowledge or assumptions, which can limit the diversity of source domain data. In this study, we propose a novel random feature augmentation (RFA) method to diversify source domain data at the feature level without prior knowledge. Specifically, our RFA method perturbs domain-specific information while preserving domain-invariant information, thereby adequately diversifying the source domain data. Furthermore, we propose a dual-branches invariant synergistic learning strategy to capture domain-invariant information from the augmented features of RFA, enabling DCNNs to learn a more generalized representation. We evaluate our proposed method on two challenging medical image segmentation tasks, optic cup/disc segmentation on fundus images and prostate segmentation on MRI images. Extensive experimental results demonstrate the superior performance of our method over state-of-the-art domain generalization methods.

A deep learning framework for identifying and segmenting three vessels in fetal heart ultrasound images.

BACKGROUND: Congenital heart disease (CHD) is one of the most common birth defects in the world. It is the leading cause of infant mortality, necessitating an early diagnosis for timely intervention. Prenatal screening using ultrasound is the primary method for CHD detection. However, its effectiveness is heavily reliant on the expertise of physicians, leading to subjective interpretations and potential underdiagnosis. Therefore, a method for automatic analysis of fetal cardiac ultrasound images is highly desired to assist an objective and effective CHD diagnosis. METHOD: In this study, we propose a deep learning-based framework for the identification and segmentation of the three vessels-the pulmonary artery, aorta, and superior vena cava-in the ultrasound three vessel view (3VV) of the fetal heart. In the first stage of the framework, the object detection model Yolov5 is employed to identify the three vessels and localize the Region of Interest (ROI) within the original full-sized ultrasound images. Subsequently, a modified Deeplabv3 equipped with our novel AMFF (Attentional Multi-scale Feature Fusion) module is applied in the second stage to segment the three vessels within the cropped ROI images. RESULTS: We evaluated our method with a dataset consisting of 511 fetal heart 3VV images. Compared to existing models, our framework exhibits superior performance in the segmentation of all the three vessels, demonstrating the Dice coefficients of 85.55%, 89.12%, and 77.54% for PA, Ao and SVC respectively. CONCLUSIONS: Our experimental results show that our proposed framework can automatically and accurately detect and segment the three vessels in fetal heart 3VV images. This method has the potential to assist sonographers in enhancing the precision of vessel assessment during fetal heart examinations.

Medical image segmentation of gastric adenocarcinoma based on dense connection of residuals.

BACKGROUND AND OBJECTIVE: Accurate segmentation of gastric cancer based on CT images of gastric adenocarcinoma is crucial for physicians to screen gastric diseases, clinical diagnosis, preoperative prediction, and postoperative evaluation plans. To address the issue of the inability of the segmentation algorithm to depict the correct boundaries due to unclear gastric contours in the lesion area and the visible irregular band-like dense shadow extending to the perigastric region, a 3D medical image segmentation model 3D UNet based on residual dense jumping method is proposed. METHODS: In the method we proposed, Residual Dense Block, which is applied to the image super-resolution module to remove CT artifacts, and Residual Block in ResNet are further fused. The quality of CT images is improved by Residual Dense Skip Block, which removes banded dense shadows, preserves image details and edge information, captures features, and improves the segmentation performance of gastric adenocarcinoma. The Instance Normalization layer position is modified to select the best result. Different loss functions are also combined in order to obtain the best gastric adenocarcinoma segmentation performance. RESULTS: We tested the model on a hospital-provided gastric adenocarcinoma dataset. The experimental results show that our model outperforms the existing methods in CT gastric adenocarcinoma segmentation, in which the method combining the hybrid loss function of Dice and CE obtains an average dice score of 82.3%, which is improved by 5.3% and 3.8% compared to TransUNet and Hiformer, respectively, and improves the cross-merge rate to 70.8%, compared to nnFormer, nnUNet by 1% and 0.9%, respectively. CONCLUSIONS: The residual jump connection structure indeed improves segmentation performance. The proposed method has the potential to be used as a screen for gastric diseases and to assist physicians in diagnosis.

A Novel Tongue Coating Segmentation Method Based on Improved TransUNet.

BACKGROUND: As an important part of the tongue, the tongue coating is closely associated with different disorders and has major diagnostic benefits. This study aims to construct a neural network model that can perform complex tongue coating segmentation. This addresses the issue of tongue coating segmentation in intelligent tongue diagnosis automation. METHOD: This work proposes an improved TransUNet to segment the tongue coating. We introduced a transformer as a self-attention mechanism to capture the semantic information in the high-level features of the encoder. At the same time, the subtraction feature pyramid (SFP) and visual regional enhancer (VRE) were constructed to minimize the redundant information transmitted by skip connections and improve the spatial detail information in the low-level features of the encoder. RESULTS: Comparative and ablation experimental findings indicate that our model has an accuracy of 96.36%, a precision of 96.26%, a dice of 96.76%, a recall of 97.43%, and an IoU of 93.81%. Unlike the reference model, our model achieves the best segmentation effect. CONCLUSION: The improved TransUNet proposed here can achieve precise segmentation of complex tongue images. This provides an effective technique for the automatic extraction in images of the tongue coating, contributing to the automation and accuracy of tongue diagnosis.

A Retinal Vessel Segmentation Method Based on the Sharpness-Aware Minimization Model.

Retinal vessel segmentation is crucial for diagnosing and monitoring various eye diseases such as diabetic retinopathy, glaucoma, and hypertension. In this study, we examine how sharpness-aware minimization (SAM) can improve RF-UNet's generalization performance. RF-UNet is a novel model for retinal vessel segmentation. We focused our experiments on the digital retinal images for vessel extraction (DRIVE) dataset, which is a benchmark for retinal vessel segmentation, and our test results show that adding SAM to the training procedure leads to notable improvements. Compared to the non-SAM model (training loss of 0.45709 and validation loss of 0.40266), the SAM-trained RF-UNet model achieved a significant reduction in both training loss (0.094225) and validation loss (0.08053). Furthermore, compared to the non-SAM model (training accuracy of 0.90169 and validation accuracy of 0.93999), the SAM-trained model demonstrated higher training accuracy (0.96225) and validation accuracy (0.96821). Additionally, the model performed better in terms of sensitivity, specificity, AUC, and F1 score, indicating improved generalization to unseen data. Our results corroborate the notion that SAM facilitates the learning of flatter minima, thereby improving generalization, and are consistent with other research highlighting the advantages of advanced optimization methods. With wider implications for other medical imaging tasks, these results imply that SAM can successfully reduce overfitting and enhance the robustness of retinal vessel segmentation models. Prospective research avenues encompass verifying the model on vaster and more diverse datasets and investigating its practical implementation in real-world clinical situations.

An Automatic Method for Elbow Joint Recognition, Segmentation and Reconstruction.

Elbow computerized tomography (CT) scans have been widely applied for describing elbow morphology. To enhance the objectivity and efficiency of clinical diagnosis, an automatic method to recognize, segment, and reconstruct elbow joint bones is proposed in this study. The method involves three steps: initially, the humerus, ulna, and radius are automatically recognized based on the anatomical features of the elbow joint, and the prompt boxes are generated. Subsequently, elbow MedSAM is obtained through transfer learning, which accurately segments the CT images by integrating the prompt boxes. After that, hole-filling and object reclassification steps are executed to refine the mask. Finally, three-dimensional (3D) reconstruction is conducted seamlessly using the marching cube algorithm. To validate the reliability and accuracy of the method, the images were compared to the masks labeled by senior surgeons. Quantitative evaluation of segmentation results revealed median intersection over union (IoU) values of 0.963, 0.959, and 0.950 for the humerus, ulna, and radius, respectively. Additionally, the reconstructed surface errors were measured at 1.127, 1.523, and 2.062 mm, respectively. Consequently, the automatic elbow reconstruction method demonstrates promising capabilities in clinical diagnosis, preoperative planning, and intraoperative navigation for elbow joint diseases.

NTSM: a non-salient target segmentation model for oral mucosal diseases.

BACKGROUND: Oral mucosal diseases are similar to the surrounding normal tissues, i.e., their many non-salient features, which poses a challenge for accurate segmentation lesions. Additionally, high-precision large models generate too many parameters, which puts pressure on storage and makes it difficult to deploy on portable devices. METHODS: To address these issues, we design a non-salient target segmentation model (NTSM) to improve segmentation performance while reducing the number of parameters. The NTSM includes a difference association (DA) module and multiple feature hierarchy pyramid attention (FHPA) modules. The DA module enhances feature differences at different levels to learn local context information and extend the segmentation mask to potentially similar areas. It also learns logical semantic relationship information through different receptive fields to determine the actual lesions and further elevates the segmentation performance of non-salient lesions. The FHPA module extracts pathological information from different views by performing the hadamard product attention (HPA) operation on input features, which reduces the number of parameters. RESULTS: The experimental results on the oral mucosal diseases (OMD) dataset and international skin imaging collaboration (ISIC) dataset demonstrate that our model outperforms existing state-of-the-art methods. Compared with the nnU-Net backbone, our model has 43.20% fewer parameters while still achieving a 3.14% increase in the Dice score. CONCLUSIONS: Our model has high segmentation accuracy on non-salient areas of oral mucosal diseases and can effectively reduce resource consumption.

PIS-Net: Efficient Medical Image Segmentation Network with Multivariate Downsampling for Point-of-Care.

Recently, with more portable diagnostic devices being moved to people anywhere, point-of-care (PoC) imaging has become more convenient and more popular than the traditional "bed imaging". Instant image segmentation, as an important technology of computer vision, is receiving more and more attention in PoC diagnosis. However, the image distortion caused by image preprocessing and the low resolution of medical images extracted by PoC devices are urgent problems that need to be solved. Moreover, more efficient feature representation is necessary in the design of instant image segmentation. In this paper, a new feature representation considering the relationships among local features with minimal parameters and a lower computational complexity is proposed. Since a feature window sliding along a diagonal can capture more pluralistic features, a Diagonal-Axial Multi-Layer Perceptron is designed to obtain the global correlation among local features for a more comprehensive feature representation. Additionally, a new multi-scale feature fusion is proposed to integrate nonlinear features with linear ones to obtain a more precise feature representation. Richer features are figured out. In order to improve the generalization of the models, a dynamic residual spatial pyramid pooling based on various receptive fields is constructed according to different sizes of images, which alleviates the influence of image distortion. The experimental results show that the proposed strategy has better performance on instant image segmentation. Notably, it yields an average improvement of 1.31% in Dice than existing strategies on the BUSI, ISIC2018 and MoNuSeg datasets.

[Medical image segmentation data augmentation method based on channel weight and data-efficient features].

In computer-aided medical diagnosis, obtaining labeled medical image data is expensive, while there is a high demand for model interpretability. However, most deep learning models currently require a large amount of data and lack interpretability. To address these challenges, this paper proposes a novel data augmentation method for medical image segmentation. The uniqueness and advantages of this method lie in the utilization of gradient-weighted class activation mapping to extract data efficient features, which are then fused with the original image. Subsequently, a new channel weight feature extractor is constructed to learn the weights between different channels. This approach achieves non-destructive data augmentation effects, enhancing the model's performance, data efficiency, and interpretability. Applying the method of this paper to the Hyper-Kvasir dataset, the intersection over union (IoU) and Dice of the U-net were improved, respectively; and on the ISIC-Archive dataset, the IoU and Dice of the DeepLabV3+ were also improved respectively. Furthermore, even when the training data is reduced to 70 %, the proposed method can still achieve performance that is 95 % of that achieved with the entire dataset, indicating its good data efficiency. Moreover, the data-efficient features used in the method have interpretable information built-in, which enhances the interpretability of the model. The method has excellent universality, is plug-and-play, applicable to various segmentation methods, and does not require modification of the network structure, thus it is easy to integrate into existing medical image segmentation method, enhancing the convenience of future research and applications.

[Multi-scale medical image segmentation based on pixel encoding and spatial attention mechanism].

In response to the issues of single-scale information loss and large model parameter size during the sampling process in U-Net and its variants for medical image segmentation, this paper proposes a multi-scale medical image segmentation method based on pixel encoding and spatial attention. Firstly, by redesigning the input strategy of the Transformer structure, a pixel encoding module is introduced to enable the model to extract global semantic information from multi-scale image features, obtaining richer feature information. Additionally, deformable convolutions are incorporated into the Transformer module to accelerate convergence speed and improve module performance. Secondly, a spatial attention module with residual connections is introduced to allow the model to focus on the foreground information of the fused feature maps. Finally, through ablation experiments, the network is lightweighted to enhance segmentation accuracy and accelerate model convergence. The proposed algorithm achieves satisfactory results on the Synapse dataset, an official public dataset for multi-organ segmentation provided by the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), with Dice similarity coefficient (DSC) and 95% Hausdorff distance (HD95) scores of 77.65 and 18.34, respectively. The experimental results demonstrate that the proposed algorithm can enhance multi-organ segmentation performance, potentially filling the gap in multi-scale medical image segmentation algorithms, and providing assistance for professional physicians in diagnosis.

P-TransUNet: an improved parallel network for medical image segmentation.

Deep learning-based medical image segmentation has made great progress over the past decades. Scholars have proposed many novel transformer-based segmentation networks to solve the problems of building long-range dependencies and global context connections in convolutional neural networks (CNNs). However, these methods usually replace the CNN-based blocks with improved transformer-based structures, which leads to the lack of local feature extraction ability, and these structures require a huge number of data for training. Moreover, those methods did not pay attention to edge information, which is essential in medical image segmentation. To address these problems, we proposed a new network structure, called P-TransUNet. This network structure combines the designed efficient P-Transformer and the fusion module, which extract distance-related long-range dependencies and local information respectively and produce the fused features. Besides, we introduced edge loss into training to focus the attention of the network on the edge of the lesion area to improve segmentation performance. Extensive experiments across four tasks of medical image segmentation demonstrated the effectiveness of P-TransUNet, and showed that our network outperforms other state-of-the-art methods.

EG-TransUNet: a transformer-based U-Net with enhanced and guided models for biomedical image segmentation.

Although various methods based on convolutional neural networks have improved the performance of biomedical image segmentation to meet the precision requirements of medical imaging segmentation task, medical image segmentation methods based on deep learning still need to solve the following problems: (1) Difficulty in extracting the discriminative feature of the lesion region in medical images during the encoding process due to variable sizes and shapes; (2) difficulty in fusing spatial and semantic information of the lesion region effectively during the decoding process due to redundant information and the semantic gap. In this paper, we used the attention-based Transformer during the encoder and decoder stages to improve feature discrimination at the level of spatial detail and semantic location by its multihead-based self-attention. In conclusion, we propose an architecture called EG-TransUNet, including three modules improved by a transformer: progressive enhancement module, channel spatial attention, and semantic guidance attention. The proposed EG-TransUNet architecture allowed us to capture object variabilities with improved results on different biomedical datasets. EG-TransUNet outperformed other methods on two popular colonoscopy datasets (Kvasir-SEG and CVC-ClinicDB) by achieving 93.44% and 95.26% on mDice. Extensive experiments and visualization results demonstrate that our method advances the performance on five medical segmentation datasets with better generalization ability.

LSW-Net: Lightweight Deep Neural Network Based on Small-World properties for Spine MR Image Segmentation.

BACKGROUND: Segmenting spinal tissues from MR images is important for automatic image analysis. Deep neural network-based segmentation methods are efficient, yet have high computational costs. PURPOSE: To design a lightweight model based on small-world properties (LSW-Net) to segment spinal MR images, suitable for low-computing-power embedded devices. STUDY TYPE: Retrospective. POPULATION: A total of 386 subjects (2948 images) from two independent sources. Dataset I: 214 subjects/779 images, all for disk degeneration screening, 147 had disk degeneration, 52 had herniated disc. Dataset II: 172 subjects/2169 images, 142 patients with vertebral degeneration, 163 patients with disc degeneration. 70% images in each dataset for training, 20% for validation, and 10% for testing. FIELD STRENGTH/SEQUENCE: T1- and T2-weighted turbo spin echo sequences at 3 T. ASSESSMENT: Segmentation performance of LSW-Net was compared with four mainstream (including U-net and U-net++) and five lightweight models using five radiologists' manual segmentations (vertebrae, disks, spinal fluid) as reference standard. LSW-Net was also deployed on NVIDIA Jetson nano to compare the pixels number in segmented vertebrae and disks. STATISTICAL TESTS: All models were evaluated with accuracy, precision, Dice similarity coefficient (DSC), and area under the receiver operating characteristic (AUC). Pixel numbers segmented by LSW-Net on the embedded device were compared with manual segmentation using paired t-tests, with P < 0.05 indicating significance. RESULTS: LSW-Net had 98.5% fewer parameters than U-net but achieved similar accuracy in both datasets (dataset I: DSC 0.84 vs. 0.87, AUC 0.92 vs. 0.94; dataset II: DSC 0.82 vs. 0.82, AUC 0.88 vs. 0.88). LSW-Net showed no significant differences in pixel numbers for vertebrae (dataset I: 5893.49 vs. 5752.61, P = 0.21; dataset II: 5073.42 vs. 5137.12, P = 0.56) and disks (dataset I: 1513.07 vs. 1535.69, P = 0.42; dataset II: 1049.74 vs. 1087.88, P = 0.24) segmentation on an embedded device compared to manual segmentation. DATA CONCLUSION: Proposed LSW-Net achieves high accuracy with fewer parameters than U-net and can be deployed on embedded device, facilitating wider application. EVIDENCE LEVEL: 2. TECHNICAL EFFICACY: 1.

MRDFF: A deep forest based framework for CT whole heart segmentation.

Automatic whole heart segmentation plays an important role in the treatment and research of cardiovascular diseases. In this paper, we propose an improved Deep Forest framework, named Multi-Resolution Deep Forest Framework (MRDFF), which accomplishes whole heart segmentation in two stages. We extract the heart region by binary classification in the first stage, thus avoiding the class imbalance problem caused by too much background. The results of the first stage are then subdivided in the second stage to obtain accurate cardiac substructures. In addition, we also propose hybrid feature fusion, multi-resolution fusion and multi-scale fusion to further improve the segmentation accuracy. Experiments on the public dataset MM-WHS show that our model can achieve comparable accuracy in about half the training time of neural network models.

2D medical image segmentation via learning multi-scale contextual dependencies.

Automatic medical image segmentation plays an important role as a diagnostic aid in the identification of diseases and their treatment in clinical settings. Recently proposed methods based on Convolutional Neural Networks (CNNs) have demonstrated their potential in image processing tasks, including some medical image analysis tasks. Those methods can learn various feature representations with numerous weight-shared convolutional kernels, however, the missed diagnosis rate of regions of interest (ROIs) is still high in medical image segmentation. Two crucial factors behind this shortcoming, which have been overlooked, are small ROIs from medical images and the limited context information from the existing network models. In order to reduce the missed diagnosis rate of ROIs from medical images, we propose a new segmentation framework which enhances the representative capability of small ROIs (particularly in deep layers) and explicitly learns global contextual dependencies in multi-scale feature spaces. In particular, the local features and their global dependencies from each feature space are adaptively aggregated based on both the spatial and the channel dimensions. Moreover, some visualization comparisons of the learned features from our framework further boost neural networks' interpretability. Experimental results show that, in comparison to some popular medical image segmentation and general image segmentation methods, our proposed framework achieves the state-of-the-art performance on the liver tumor segmentation task with 91.18% Sensitivity, the COVID-19 lung infection segmentation task with 75.73% Sensitivity and the retinal vessel detection task with 82.68% Sensitivity. Moreover, it is possible to integrate (parts of) the proposed framework into most of the recently proposed Fully CNN-based models, in order to improve their effectiveness in medical image segmentation tasks.

Fovea-UNet: detection and segmentation of lymph node metastases in colorectal cancer with deep learning.

BACKGROUND: Colorectal cancer is one of the most serious malignant tumors, and lymph node metastasis (LNM) from colorectal cancer is a major factor for patient management and prognosis. Accurate image detection of LNM is an important task to help clinicians diagnose cancer. Recently, the U-Net architecture based on convolutional neural networks (CNNs) has been widely used to segment image to accomplish more precise cancer diagnosis. However, the accurate segmentation of important regions with high diagnostic value is still a great challenge due to the insufficient capability of CNN and codec structure in aggregating the detailed and non-local contextual information. In this work, we propose a high performance and low computation solution. METHODS: Inspired by the working principle of Fovea in visual neuroscience, a novel network framework based on U-Net for cancer segmentation named Fovea-UNet is proposed to adaptively adjust the resolution according to the importance-aware of information and selectively focuses on the region most relevant to colorectal LNM. Specifically, we design an effective adaptively optimized pooling operation called Fovea Pooling (FP), which dynamically aggregate the detailed and non-local contextual information according to the pixel-level feature importance. In addition, the improved lightweight backbone network based on GhostNet is adopted to reduce the computational cost caused by FP. RESULTS: Experimental results show that our proposed framework can achieve higher performance than other state-of-the-art segmentation networks with 79.38% IoU, 88.51% DSC, 92.82% sensitivity and 84.57% precision on the LNM dataset, and the parameter amount is reduced to 23.23 MB. CONCLUSIONS: The proposed framework can provide a valid tool for cancer diagnosis, especially for LNM of colorectal cancer.

Swin Unet3D: a three-dimensional medical image segmentation network combining vision transformer and convolution.

BACKGROUND: Semantic segmentation of brain tumors plays a critical role in clinical treatment, especially for three-dimensional (3D) magnetic resonance imaging, which is often used in clinical practice. Automatic segmentation of the 3D structure of brain tumors can quickly help physicians understand the properties of tumors, such as the shape and size, thus improving the efficiency of preoperative planning and the odds of successful surgery. In past decades, 3D convolutional neural networks (CNNs) have dominated automatic segmentation methods for 3D medical images, and these network structures have achieved good results. However, to reduce the number of neural network parameters, practitioners ensure that the size of convolutional kernels in 3D convolutional operations generally does not exceed [Formula: see text], which also leads to CNNs showing limitations in learning long-distance dependent information. Vision Transformer (ViT) is very good at learning long-distance dependent information in images, but it suffers from the problems of many parameters. What's worse, the ViT cannot learn local dependency information in the previous layers under the condition of insufficient data. However, in the image segmentation task, being able to learn this local dependency information in the previous layers makes a big impact on the performance of the model. METHODS: This paper proposes the Swin Unet3D model, which represents voxel segmentation on medical images as a sequence-to-sequence prediction. The feature extraction sub-module in the model is designed as a parallel structure of Convolution and ViT so that all layers of the model are able to adequately learn both global and local dependency information in the image. RESULTS: On the validation dataset of Brats2021, our proposed model achieves dice coefficients of 0.840, 0.874, and 0.911 on the ET channel, TC channel, and WT channel, respectively. On the validation dataset of Brats2018, our model achieves dice coefficients of 0.716, 0.761, and 0.874 on the corresponding channels, respectively. CONCLUSION: We propose a new segmentation model that combines the advantages of Vision Transformer and Convolution and achieves a better balance between the number of model parameters and segmentation accuracy. The code can be found at https://github.com/1152545264/SwinUnet3D .

2.5D MFFAU-Net: a convolutional neural network for kidney segmentation.

BACKGROUND: Kidney tumors have become increasingly prevalent among adults and are now considered one of the most common types of tumors. Accurate segmentation of kidney tumors can help physicians assess tumor complexity and aggressiveness before surgery. However, segmenting kidney tumors manually can be difficult because of their heterogeneity. METHODS: This paper proposes a 2.5D MFFAU-Net (multi-level Feature Fusion Attention U-Net) to segment kidneys, tumors and cysts. First, we propose a 2.5D model for learning to combine and represent a given slice in 2D slices, thereby introducing 3D information to balance memory consumption and model complexity. Then, we propose a ResConv architecture in MFFAU-Net and use the high-level and low-level feature in the model. Finally, we use multi-level information to analyze the spatial features between slices to segment kidneys and tumors. RESULTS: The 2.5D MFFAU-Net was evaluated on KiTS19 and KiTS21 kidney datasets and demonstrated an average dice score of 0.924 and 0.875, respectively, and an average Surface dice (SD) score of 0.794 in KiTS21. CONCLUSION: The 2.5D MFFAU-Net model can effectively segment kidney tumors, and the results are comparable to those obtained with high-performance 3D CNN models, and have the potential to serve as a point of reference in clinical practice.