A lightweight dual-path cascaded network for vessel segmentation in fundus image.

Automatic and fast segmentation of retinal vessels in fundus images is a prerequisite in clinical ophthalmic diseases; however, the high model complexity and low segmentation accuracy still limit its application. This paper proposes a lightweight dual-path cascaded network (LDPC-Net) for automatic and fast vessel segmentation. We designed a dual-path cascaded network via two U-shaped structures. Firstly, we employed a structured discarding (SD) convolution module to alleviate the over-fitting problem in both codec parts. Secondly, we introduced the depthwise separable convolution (DSC) technique to reduce the parameter amount of the model. Thirdly, a residual atrous spatial pyramid pooling (ResASPP) model is constructed in the connection layer to aggregate multi-scale information effectively. Finally, we performed comparative experiments on three public datasets. Experimental results show that the proposed method achieved superior performance on the accuracy, connectivity, and parameter quantity, thus proving that it can be a promising lightweight assisted tool for ophthalmic diseases.

MIC-Net: multi-scale integrated context network for automatic retinal vessel segmentation in fundus image.

PURPOSE: Accurate retinal vessel segmentation is of great value in the auxiliary screening of various diseases. However, due to the low contrast between the ends of the branches of the fundus blood vessels and the background, and the variable morphology of the optic disc and cup in the retinal image, the task of high-precision retinal blood vessel segmentation still faces difficulties. METHOD: This paper proposes a multi-scale integrated context network, MIC-Net, which fully fuses the encoder-decoder features, and extracts multi-scale information. First, a hybrid stride sampling (HSS) block was designed in the encoder to minimize the loss of helpful information caused by the downsampling operation. Second, a dense hybrid dilated convolution (DHDC) was employed in the connection layer. On the premise of preserving feature resolution, it can perceive richer contextual information. Third, a squeeze-and-excitation with residual connections (SERC) was introduced in the decoder to adjust the channel attention adaptively. Finally, we utilized a multi-layer feature fusion mechanism in the skip connection part, which enables the network to consider both low-level details and high-level semantic information. RESULTS: We evaluated the proposed method on three public datasets DRIVE, STARE and CHASE. In the experimental results, the Area under the receiver operating characteristic (ROC) and the accuracy rate (Acc) achieved high performances of 98.62%/97.02%, 98.60%/97.76% and 98.73%/97.38%, respectively. CONCLUSIONS: Experimental results show that the proposed method can obtain comparable segmentation performance compared with the state-of-the-art (SOTA) methods. Specifically, the proposed method can effectively reduce the small blood vessel segmentation error, thus proving it a promising tool for auxiliary diagnosis of ophthalmic diseases.

Multi-scale attention and deep supervision-based 3D UNet for automatic liver segmentation from CT.

BACKGROUND: Automatic liver segmentation is a prerequisite for hepatoma treatment; however, the low accuracy and stability hinder its clinical application. To alleviate this limitation, we deeply mine the context information of different scales and combine it with deep supervision to improve the accuracy of liver segmentation in this paper. METHODS: We proposed a new network called MAD-UNet for automatic liver segmentation from CT. It is grounded in the 3D UNet and leverages multi-scale attention and deep supervision mechanisms. In the encoder, the downsampling pooling in 3D UNet is replaced by convolution to alleviate the loss of feature information. Meanwhile, the residual module is introduced to avoid gradient vanishment. Besides, we use the long-short skip connections (LSSC) to replace the ordinary skip connections to preserve more edge detail. In the decoder, the features of different scales are aggregated, and the attention module is employed to capture the spatial context information. Moreover, we utilized the deep supervision mechanism to improve the learning ability on deep and shallow information. RESULTS: We evaluated the proposed method on three public datasets, including, LiTS17, SLiver07, and 3DIRCADb, and obtained Dice scores of 0.9727, 0.9752, and 0.9691 for liver segmentation, respectively, which outperform the other state-of-the-art (SOTA) methods. CONCLUSIONS: Both qualitative and quantitative experimental results demonstrate that the proposed method can make full use of the feature information of different stages while enhancing spatial data's learning ability, thereby achieving high liver segmentation accuracy. Thus, it proved to be a promising tool for automatic liver segmentation in clinical assistance.

Deep supervision and atrous inception-based U-Net combining CRF for automatic liver segmentation from CT.

Due to low contrast and the blurred boundary between liver tissue and neighboring organs sharing similar intensity values, the problem of liver segmentation from CT images has not yet achieved satisfactory performance and remains a challenge. To alleviate these problems, we introduce deep supervision (DS) and atrous inception (AI) technologies with conditional random field (CRF) and propose three major improvements that are experimentally shown to have substantive and practical value. First, we replace the encoder's standard convolution with the residual block. Residual blocks can increase the depth of the network. Second, we provide an AI module to connect the encoder and decoder. AI allows us to obtain multi-scale features. Third, we incorporate the DS mechanism into the decoder. This helps to make full use of information of the shallow layers. In addition, we employ the Tversky loss function to balance the segmented and non-segmented regions and perform further refinement with a dense CRF. Finally, we extensively validate the proposed method on three public databases: LiTS17, 3DIRCADb, and SLiver07. Compared to the state-of-the-art methods, the proposed method achieved increased segmentation accuracy for the livers with low contrast and the fuzzy boundary between liver tissue and neighboring organs and is, therefore, more suited for automatic segmentation of these livers.

SERR-U-Net: Squeeze-and-Excitation Residual and Recurrent Block-Based U-Net for Automatic Vessel Segmentation in Retinal Image.

METHODS: A new SERR-U-Net framework for retinal vessel segmentation is proposed, which leverages technologies including Squeeze-and-Excitation (SE), residual module, and recurrent block. First, the convolution layers of encoder and decoder are modified on the basis of U-Net, and the recurrent block is used to increase the network depth. Second, the residual module is utilized to alleviate the vanishing gradient problem. Finally, to derive more specific vascular features, we employed the SE structure to introduce attention mechanism into the U-shaped network. In addition, enhanced super-resolution generative adversarial networks (ESRGANs) are also deployed to remove the noise of retinal image. RESULTS: The effectiveness of this method was tested on two public datasets, DRIVE and STARE. In the experiment of DRIVE dataset, the accuracy and AUC (area under the curve) of our method were 0.9552 and 0.9784, respectively, and for SATRE dataset, 0.9796 and 0.9859 were achieved, respectively, which proved a high accuracy and promising prospect on clinical assistance. CONCLUSION: An improved U-Net network combining SE, ResNet, and recurrent technologies is developed for automatic vessel segmentation from retinal image. This new model enables an improvement on the accuracy compared to learning-based methods, and its robustness in circumvent challenging cases such as small blood vessels and intersection of vessels is also well demonstrated and validated.

SAR-U-Net: Squeeze-and-excitation block and atrous spatial pyramid pooling based residual U-Net for automatic liver segmentation in Computed Tomography.

BACKGROUND AND OBJECTIVE: Liver segmentation is an essential prerequisite for liver cancer diagnosis and surgical planning. Traditionally, liver contour is delineated manually by radiologist in a slice-by-slice fashion. However, this process is time-consuming and prone to errors depending on radiologist's experience. In this paper, a modified U-Net based framework is presented, which leverages techniques from Squeeze-and-Excitation (SE) block, Atrous Spatial Pyramid Pooling (ASPP) and residual learning for accurate and robust liver Computed Tomography (CT) segmentation, and the effectiveness of the proposed method was tested on two public datasets LiTS17 and SLiver07. METHODS: A new network architecture, called SAR-U-Net was designed, which is grounded in the classical U-Net. Firstly, the SE block is introduced to adaptively extract image features after each convolution in the U-Net encoder, while suppressing irrelevant regions, and highlighting features of specific segmentation task; Secondly, the ASPP is employed to replace the transition layer and the output layer, and acquire multi-scale image information via different receptive fields. Thirdly, to alleviate the gradient vanishment problem, the traditional convolution block is replaced with the residual structures, and thus prompt the network to gain accuracy from considerably increased depth. RESULTS: In the LiTS17 database experiment, five popular metrics were used for evaluation, including Dice coefficient, VOE, RVD, ASD and MSD. Compared with other closely related models, the proposed method achieved the highest accuracy. In addition, in the experiment of the SLiver07 dataset, compared with other closely related models, the proposed method achieved the highest segmentation accuracy except for the RVD. CONCLUSION: An improved U-Net network combining SE, ASPP, and residual structures is developed for automatic liver segmentation from CT images. This new model shows a great improvement on the accuracy compared to other closely related models, and its robustness to challenging problems, including small liver regions, discontinuous liver regions, and fuzzy liver boundaries, is also well demonstrated and validated.

Multi-slice low-rank tensor decomposition based multi-atlas segmentation: Application to automatic pathological liver CT segmentation.

Liver segmentation from abdominal CT images is an essential step for liver cancer computer-aided diagnosis and surgical planning. However, both the accuracy and robustness of existing liver segmentation methods cannot meet the requirements of clinical applications. In particular, for the common clinical cases where the liver tissue contains major pathology, current segmentation methods show poor performance. In this paper, we propose a novel low-rank tensor decomposition (LRTD) based multi-atlas segmentation (MAS) framework that achieves accurate and robust pathological liver segmentation of CT images. Firstly, we propose a multi-slice LRTD scheme to recover the underlying low-rank structure embedded in 3D medical images. It performs the LRTD on small image segments consisting of multiple consecutive image slices. Then, we present an LRTD-based atlas construction method to generate tumor-free liver atlases that mitigates the performance degradation of liver segmentation due to the presence of tumors. Finally, we introduce an LRTD-based MAS algorithm to derive patient-specific liver atlases for each test image, and to achieve accurate pairwise image registration and label propagation. Extensive experiments on three public databases of pathological liver cases validate the effectiveness of the proposed method. Both qualitative and quantitative results demonstrate that, in the presence of major pathology, the proposed method is more accurate and robust than state-of-the-art methods.

Automatic construction of statistical shape models using deformable simplex meshes with vector field convolution energy.

BACKGROUND: In the active shape model framework, principal component analysis (PCA) based statistical shape models (SSMs) are widely employed to incorporate high-level a priori shape knowledge of the structure to be segmented to achieve robustness. A crucial component of building SSMs is to establish shape correspondence between all training shapes, which is a very challenging task, especially in three dimensions. METHODS: We propose a novel mesh-to-volume registration based shape correspondence establishment method to improve the accuracy and reduce the computational cost. Specifically, we present a greedy algorithm based deformable simplex mesh that uses vector field convolution as the external energy. Furthermore, we develop an automatic shape initialization method by using a Gaussian mixture model based registration algorithm, to derive an initial shape that has high overlap with the object of interest, such that the deformable models can then evolve more locally. We apply the proposed deformable surface model to the application of femur statistical shape model construction to illustrate its accuracy and efficiency. RESULTS: Extensive experiments on ten femur CT scans show that the quality of the constructed femur shape models via the proposed method is much better than that of the classical spherical harmonics (SPHARM) method. Moreover, the proposed method achieves much higher computational efficiency than the SPHARM method. CONCLUSIONS: The experimental results suggest that our method can be employed for effective statistical shape model construction.

Low-rank and sparse decomposition based shape model and probabilistic atlas for automatic pathological organ segmentation.

One major limiting factor that prevents the accurate delineation of human organs has been the presence of severe pathology and pathology affecting organ borders. Overcoming these limitations is exactly what we are concerned in this study. We propose an automatic method for accurate and robust pathological organ segmentation from CT images. The method is grounded in the active shape model (ASM) framework. It leverages techniques from low-rank and sparse decomposition (LRSD) theory to robustly recover a subspace from grossly corrupted data. We first present a population-specific LRSD-based shape prior model, called LRSD-SM, to handle non-Gaussian gross errors caused by weak and misleading appearance cues of large lesions, complex shape variations, and poor adaptation to the finer local details in a unified framework. For the shape model initialization, we introduce a method based on patient-specific LRSD-based probabilistic atlas (PA), called LRSD-PA, to deal with large errors in atlas-to-target registration and low likelihood of the target organ. Furthermore, to make our segmentation framework more efficient and robust against local minima, we develop a hierarchical ASM search strategy. Our method is tested on the SLIVER07 database for liver segmentation competition, and ranks 3rd in all the published state-of-the-art automatic methods. Our method is also evaluated on some pathological organs (pathological liver and right lung) from 95 clinical CT scans and its results are compared with the three closely related methods. The applicability of the proposed method to segmentation of the various pathological organs (including some highly severe cases) is demonstrated with good results on both quantitative and qualitative experimentation; our segmentation algorithm can delineate organ boundaries that reach a level of accuracy comparable with those of human raters.