1
|
Liu J, Zhao D, Shen J, Geng P, Zhang Y, Yang J, Zhang Z. HRD-Net: High resolution segmentation network with adaptive learning ability of retinal vessel features. Comput Biol Med 2024; 173:108295. [PMID: 38520920 DOI: 10.1016/j.compbiomed.2024.108295] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 12/06/2023] [Revised: 01/31/2024] [Accepted: 03/12/2024] [Indexed: 03/25/2024]
Abstract
Retinal segmentation is a crucial step in the early warning of human health conditions. However, retinal blood vessels possess complex curvature, irregular distribution, and contain multi-scale fine structures, which make the limited receptive field of regular convolution challenging to process their vascular details efficiently. Additionally, the encoder-decoder based network leads to irreversible spatial information loss because of multiple downsampling, resulting in over-segmentation and missed segmentation of the vessels. For this reason, we develop a high-resolution network based on Deformable Convolution v3, called HRD-Net. By constructing a high-resolution representation, the network allows special attention to be paid to the details of tiny blood vessels. The proposed feature enhancement cascade module based on Deformable Convolution v3 can flexibly adapt and capture the ever-changing morphology and intricate connections of retinal blood vessels, ensuring the continuity of vessel segmentation. In the output phase of the network, the proposed global aggregation module integrates full-resolution feature maps while suppressing redundant features, achieving an effective fusion of high-level semantic information and spatial detail information. In addition, we have re-examined the selection criteria for activation and normalization methods, and also refine the network architectures from a spatial domain perspective to release redundant computational loads. Testing on the DRIVE, STARE, and CHASE_DB1 datasets indicates that HRD-Net, with fewer parameters, outperforms existing segmentation methods on several evaluation metrics such as F1, ACC, SE, SP, AUC, and IOU.
Collapse
Affiliation(s)
- Jianhua Liu
- School of Electrical and Electronic Engineering, Shijiazhuang Tiedao University, Shijiazhuang, 050043, China; Hebei Provincial Collaborative Innovation Center of Transportation Power Grid Intelligent Integration Technology and Equipment, School of Electrical and Electronic Engineering, Shijiazhuang Tiedao University, Shijiazhuang, China.
| | - Dongxin Zhao
- School of Electrical and Electronic Engineering, Shijiazhuang Tiedao University, Shijiazhuang, 050043, China.
| | - Juncai Shen
- College of Information Science and Technology, Shijiazhuang Tiedao University, Shijiazhuang, 050043, China.
| | - Peng Geng
- College of Information Science and Technology, Shijiazhuang Tiedao University, Shijiazhuang, 050043, China.
| | - Ying Zhang
- College of Resources and Environment, Xingtai University, Xingtai, 054001, China.
| | - Jiaxin Yang
- School of Electrical and Electronic Engineering, Shijiazhuang Tiedao University, Shijiazhuang, 050043, China.
| | - Ziqian Zhang
- School of Electrical and Electronic Engineering, Shijiazhuang Tiedao University, Shijiazhuang, 050043, China.
| |
Collapse
|
2
|
Fan X, Zhou J, Jiang X, Xin M, Hou L. CSAP-UNet: Convolution and self-attention paralleling network for medical image segmentation with edge enhancement. Comput Biol Med 2024; 172:108265. [PMID: 38461698 DOI: 10.1016/j.compbiomed.2024.108265] [Citation(s) in RCA: 3] [Impact Index Per Article: 3.0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 08/27/2023] [Revised: 02/14/2024] [Accepted: 03/06/2024] [Indexed: 03/12/2024]
Abstract
Convolution operation is performed within a local window of the input image. Therefore, convolutional neural network (CNN) is skilled in obtaining local information. Meanwhile, the self-attention (SA) mechanism extracts features by calculating the correlation between tokens from all positions in the image, which has advantage in obtaining global information. Therefore, the two modules can complement each other to improve feature extraction ability. An effective fusion method is a problem worthy of further study. In this paper, we propose a CNN and SA paralleling network CSAP-UNet with U-Net as backbone. The encoder consists of two parallel branches of CNN and Transformer to extract the feature from the input image, which takes into account both the global dependencies and the local information. Because medical images come from certain frequency bands within the spectrum, their color channels are not as uniform as natural images. Meanwhile, medical segmentation pays more attention to lesion regions in the image. Attention fusion module (AFM) integrates channel attention and spatial attention in series to fuse the output features of the two branches. The medical image segmentation task is essentially to locate the boundary of the object in the image. The boundary enhancement module (BEM) is designed in the shallow layer of the proposed network to focus more specifically on pixel-level edge details. Experimental results on three public datasets validate that CSAP-UNet outperforms state-of-the-art networks, particularly on the ISIC 2017 dataset. The cross-dataset evaluation on Kvasir and CVC-ClinicDB shows that CSAP-UNet has strong generalization ability. Ablation experiments also indicate the effectiveness of the designed modules. The code for training and test is available at https://github.com/zhouzhou1201/CSAP-UNet.git.
Collapse
Affiliation(s)
- Xiaodong Fan
- Faculty of Electrical and Control Engineering, Liaoning Technical University, Huludao, 125105, Liaoning, China.
| | - Jing Zhou
- College of Mathematics, Bohai University, Jinzhou, 121013, Liaoning, China
| | - Xiaoli Jiang
- College of Mathematics, Bohai University, Jinzhou, 121013, Liaoning, China
| | - Meizhuo Xin
- College of Mathematics, Bohai University, Jinzhou, 121013, Liaoning, China
| | - Limin Hou
- Faculty of Electrical and Control Engineering, Liaoning Technical University, Huludao, 125105, Liaoning, China
| |
Collapse
|
3
|
Luo X, Zhang H, Huang X, Gong H, Zhang J. DBNet-SI: Dual branch network of shift window attention and inception structure for skin lesion segmentation. Comput Biol Med 2024; 170:108090. [PMID: 38320341 DOI: 10.1016/j.compbiomed.2024.108090] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 07/18/2023] [Revised: 12/27/2023] [Accepted: 01/27/2024] [Indexed: 02/08/2024]
Abstract
The U-shaped convolutional neural network (CNN) has attained remarkable achievements in the segmentation of skin lesion. However, given the inherent locality of convolution, this architecture cannot capture long-range pixel dependencies and multiscale global contextual information effectively. Moreover, repeated convolutions and downsampling operations can readily result in the omission of intricate local fine-grained details. In this paper, we proposed a U-shaped network (DBNet-SI) equipped with a dual-branch module that combines shift window attention and inception structures. First, we proposed a dual-branch module that combines shift window attention and inception structures (MSI) to better capture multiscale global contextual information and long-range pixel dependencies. Specifically, we have devised a cross-branch bidirectional interaction module within the MSI module to enable information complementarity between the two branches in the channel and spatial dimensions. Therefore, MSI is capable of extracting distinguishing and comprehensive features to accurately identify the skin lesion boundaries. Second, we have devised a progressive feature enhancement and information compensation module (PFEIC), which progressively compensates for fine-grained features through reconstructed skip connections and integrated global context attention modules. The results of the experiment show the superior segmentation performance of DBNet-SI compared with other deep learning models for skin lesion segmentation in the ISIC2017 and ISIC2018 datasets. Ablation studies demonstrate that our model can effectively extract rich multiscale global contextual information and compensate for the loss of local details.
Collapse
Affiliation(s)
- Xuqiong Luo
- School of Mathematics and Statistics, Changsha University of Science and Technology, ChangSha 410114, China
| | - Hao Zhang
- School of Mathematics and Statistics, Changsha University of Science and Technology, ChangSha 410114, China
| | - Xiaofei Huang
- School of Mathematics and Statistics, Changsha University of Science and Technology, ChangSha 410114, China
| | - Hongfang Gong
- School of Mathematics and Statistics, Changsha University of Science and Technology, ChangSha 410114, China.
| | - Jin Zhang
- School of Computer and Communication Engineering, Changsha University of Science and Technology, ChangSha 410114, China
| |
Collapse
|
4
|
Peng K, Li Y, Xia Q, Liu T, Shi X, Chen D, Li L, Zhao H, Xiao H. MSMCNet: Differential context drives accurate localization and edge smoothing of lesions for medical image segmentation. Comput Biol Med 2023; 167:107624. [PMID: 37922605 DOI: 10.1016/j.compbiomed.2023.107624] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 07/07/2023] [Revised: 09/24/2023] [Accepted: 10/23/2023] [Indexed: 11/07/2023]
Abstract
Medical image segmentation plays a crucial role in clinical assistance for diagnosis. The UNet-based network architecture has achieved tremendous success in the field of medical image segmentation. However, most methods commonly employ element-wise addition or channel merging to fuse features, resulting in smaller differentiation of feature information and excessive redundancy. Consequently, this leads to issues such as inaccurate lesion localization and blurred boundaries in segmentation. To alleviate these problems, the Multi-scale Subtraction and Multi-key Context Conversion Networks (MSMCNet) are proposed for medical image segmentation. Through the construction of differentiated contextual representations, MSMCNet emphasizes vital information and achieves precise medical image segmentation by accurately localizing lesions and enhancing boundary perception. Specifically, the construction of differentiated contextual representations is accomplished through the proposed Multi-scale Non-crossover Subtraction (MSNS) module and Multi-key Context Conversion Module (MCCM). The MSNS module utilizes the context of MCCM coding and redistribute the value of feature map pixels. Extensive experiments were conducted on widely used public datasets, including the ISIC-2018 dataset, COVID-19-CT-Seg dataset, Kvasir dataset, as well as a privately constructed traumatic brain injury dataset. The experimental results demonstrated that our proposed MSMCNet outperforms state-of-the-art medical image segmentation methods across different evaluation metrics.
Collapse
Affiliation(s)
- Ke Peng
- College of Artificial Intelligent, Chongqing University of Technology, Chongqing 401135, China
| | - Yulin Li
- College of Artificial Intelligent, Chongqing University of Technology, Chongqing 401135, China
| | - Qingling Xia
- College of Artificial Intelligent, Chongqing University of Technology, Chongqing 401135, China; Department of Radiology, Chongqing University Cancer Hospital, School of Medicine, Chongqing University, Chongqing, 400030, China.
| | - Tianqi Liu
- College of Artificial Intelligent, Chongqing University of Technology, Chongqing 401135, China
| | - Xinyi Shi
- College of Artificial Intelligent, Chongqing University of Technology, Chongqing 401135, China
| | - Diyou Chen
- Institute for Traffic Medicine, Daping Hospital, Army Medical University, Chongqing 400042, China; Department of Radiology, Daping Hospital, Army Medical University, Chongqing 400042, China
| | - Li Li
- College of Artificial Intelligent, Chongqing University of Technology, Chongqing 401135, China
| | - Hui Zhao
- Institute for Traffic Medicine, Daping Hospital, Army Medical University, Chongqing 400042, China
| | - Hanguang Xiao
- College of Artificial Intelligent, Chongqing University of Technology, Chongqing 401135, China.
| |
Collapse
|
5
|
Wang K, Wang X, Xi Z, Li J, Zhang X, Wang R. Automatic Segmentation and Quantification of Abdominal Aortic Calcification in Lateral Lumbar Radiographs Based on Deep-Learning-Based Algorithms. Bioengineering (Basel) 2023; 10:1164. [PMID: 37892894 PMCID: PMC10604574 DOI: 10.3390/bioengineering10101164] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 08/07/2023] [Revised: 10/01/2023] [Accepted: 10/02/2023] [Indexed: 10/29/2023] Open
Abstract
To investigate the performance of deep-learning-based algorithms for the automatic segmentation and quantification of abdominal aortic calcification (AAC) in lateral lumbar radiographs, we retrospectively collected 1359 consecutive lateral lumbar radiographs. The data were randomly divided into model development and hold-out test datasets. The model development dataset was used to develop U-shaped fully convolutional network (U-Net) models to segment the landmarks of vertebrae T12-L5, the aorta, and anterior and posterior aortic calcifications. The AAC lengths were calculated, resulting in an automatic Kauppila score output. The vertebral levels, AAC scores, and AAC severity were obtained from clinical reports and analyzed by an experienced expert (reference standard) and the model. Compared with the reference standard, the U-Net model demonstrated a good performance in predicting the total AAC score in the hold-out test dataset, with a correlation coefficient of 0.97 (p <0.001). The overall accuracy for the AAC severity was 0.77 for the model and 0.74 for the clinical report. Additionally, the Kendall coefficient of concordance of the total AAC score prediction was 0.89 between the model-predicted score and the reference standard, and 0.88 between the structured clinical report and the reference standard. In conclusion, the U-Net-based deep learning approach demonstrated a relatively high model performance in automatically segmenting and quantifying ACC.
Collapse
Affiliation(s)
- Kexin Wang
- Department of Radiology, Peking University First Hospital, Beijing 100034, China
- School of Basic Medical Sciences, Capital Medical University, Beijing 100069, China
| | - Xiaoying Wang
- Department of Radiology, Peking University First Hospital, Beijing 100034, China
| | - Zuqiang Xi
- Beijing Smart Tree Medical Technology Co., Ltd., Beijing 102200, China
| | - Jialun Li
- Beijing Smart Tree Medical Technology Co., Ltd., Beijing 102200, China
| | - Xiaodong Zhang
- Department of Radiology, Peking University First Hospital, Beijing 100034, China
| | - Rui Wang
- Department of Radiology, Peking University First Hospital, Beijing 100034, China
| |
Collapse
|
6
|
Zang L, Liang W, Ke H, Chen F, Shen C. Research on liver cancer segmentation method based on PCNN image processing and SE-ResUnet. Sci Rep 2023; 13:12779. [PMID: 37550341 PMCID: PMC10406939 DOI: 10.1038/s41598-023-39240-0] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 03/10/2023] [Accepted: 07/21/2023] [Indexed: 08/09/2023] Open
Abstract
As one of the malignant tumors with high mortality, the initial symptoms of liver cancer are not obvious. In addition, the liver is the largest internal organ of the human body, and its structure and distribution are relatively complex. Therefore, in order to help doctors judge liver cancer more accurately, this paper proposes a variant model based on Unet network. Before segmentation, the image is preprocessed, and Pulse Coupled Neural Network (PCNN) algorithm is used to filter the image adaptively to make the image clearer. For the segmentation model, the SE module is used as the input of the residual network, and then its output is connected to the Unet model through bilinear interpolation to perform the down-sampling and up-sampling operations. The dataset is a combination of Hainan Provincial People's Hospital and some public datasets Lits. The results show that this method has better segmentation performance and accuracy than the original Unet method, and the dice coefficient, mIou and other evaluation indicators have increased by at least 2.1%, which is a method that can be applied to cancer segmentation.
Collapse
Affiliation(s)
- Lan Zang
- State Key Laboratory of Marine Resource Utilization in South China Sea, Hainan University, Haikou, 570228, China
- School of Information and Communication Engineering, Hainan University, Haikou, 570228, China
| | - Wei Liang
- State Key Laboratory of Marine Resource Utilization in South China Sea, Hainan University, Haikou, 570228, China
- School of Information and Communication Engineering, Hainan University, Haikou, 570228, China
| | - Hanchu Ke
- School of Electronic Science and Technology, Hainan University, Haikou, 570288, China.
| | - Feng Chen
- Department of Radiology, Hainan General Hospital (Hainan Affiliated Hospital of Hainan Medical University), Haikou, 570216, China.
| | - Chong Shen
- State Key Laboratory of Marine Resource Utilization in South China Sea, Hainan University, Haikou, 570228, China.
- School of Electronic Science and Technology, Hainan University, Haikou, 570288, China.
| |
Collapse
|
7
|
Fei H, Wang Q, Shang F, Xu W, Chen X, Chen Y, Li H. HC-Net: A hybrid convolutional network for non-human primate brain extraction. Front Comput Neurosci 2023; 17:1113381. [PMID: 36846727 PMCID: PMC9947775 DOI: 10.3389/fncom.2023.1113381] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 12/01/2022] [Accepted: 01/23/2023] [Indexed: 02/11/2023] Open
Abstract
Brain extraction (skull stripping) is an essential step in the magnetic resonance imaging (MRI) analysis of brain sciences. However, most of the current brain extraction methods that achieve satisfactory results for human brains are often challenged by non-human primate brains. Due to the small sample characteristics and the nature of thick-slice scanning of macaque MRI data, traditional deep convolutional neural networks (DCNNs) are unable to obtain excellent results. To overcome this challenge, this study proposed a symmetrical end-to-end trainable hybrid convolutional neural network (HC-Net). It makes full use of the spatial information between adjacent slices of the MRI image sequence and combines three consecutive slices from three axes for 3D convolutions, which reduces the calculation consumption and promotes accuracy. The HC-Net consists of encoding and decoding structures of 3D convolutions and 2D convolutions in series. The effective use of 2D convolutions and 3D convolutions relieves the underfitting of 2D convolutions to spatial features and the overfitting of 3D convolutions to small samples. After evaluating macaque brain data from different sites, the results showed that HC-Net performed better in inference time (approximately 13 s per volume) and accuracy (mean Dice coefficient reached 95.46%). The HC-Net model also had good generalization ability and stability in different modes of brain extraction tasks.
Collapse
Affiliation(s)
- Hong Fei
- College of Information and Computer, Taiyuan University of Technology, Taiyuan, China
| | - Qianshan Wang
- College of Information and Computer, Taiyuan University of Technology, Taiyuan, China
| | - Fangxin Shang
- Country Intelligent Healthcare Unit, Baidu, Beijing, China
| | - Wenyi Xu
- College of Information and Computer, Taiyuan University of Technology, Taiyuan, China
| | - Xiaofeng Chen
- College of Information and Computer, Taiyuan University of Technology, Taiyuan, China
| | - Yifei Chen
- College of Information and Computer, Taiyuan University of Technology, Taiyuan, China
| | - Haifang Li
- College of Information and Computer, Taiyuan University of Technology, Taiyuan, China,*Correspondence: Haifang Li,
| |
Collapse
|