MA-Net: A Multi-Scale Attention Network for Liver and Tumor Segmentation

Abstract

Automatic assessing the location and extent of liver and liver tumor is critical for radiologists, diagnosis and the clinical process. In recent years, a large number of variants of U-Net based on Multi-scale feature fusion are proposed to improve the segmentation performance for medical image segmentation. Unlike the previous works which extract the context information of medical image via applying the multi-scale feature fusion, we propose a novel network named Multi-scale Attention Net (MA-Net) by introducing self-attention mechanism into our method to adaptively integrate local features with their global dependencies. The MA-Net can capture rich contextual dependencies based on the attention mechanism. We design two blocks: Position-wise Attention Block (PAB) and Multi-scale Fusion Attention Block (MFAB). The PAB is used to model the feature interdependencies in spatial dimensions, which capture the spatial dependencies between pixels in a global view. In addition, the MFAB is to capture the channel dependencies between any feature map by multi-scale semantic feature fusion. We evaluate our method on the dataset of MICCAI 2017 LiTS Challenge. The proposed method achieves better performance than other state-of-the-art methods. The Dice values of liver and tumors segmentation are 0.960±0.03 and 0.749±0.08 respectively.

High Precision Mammography Lesion Identification From Imprecise Medical Annotations

“…We quantified and compared segmentation performance of our model based on Average Precision (AP) on specific IOU thresholds of 0.3, 0.5, and 0.7 as defined in the COCO ( Lin et al, 2014 ) and PASCAL VOC ( Everingham et al, 2015 ) challenges. In comparisons with U-Net ( Ronneberger et al, 2015 ) [implemented using ( Buda et al, 2019 )’s official implementation], Fully Convolutional Networks (FCN) ( Long et al, 2015 ), and Multi-Scale Attention Network (MANET) ( Fan et al, 2020 ) [implemented using ( Yakubovskiy, 2020 )] ( Figure 5 ), we found LatentCADx demonstrated the best overall AP score. The performance of AP = 0.75 was significantly better the second best model U-Net (AP = 0.62) on IOU = 0.5 threshold ( Table 1 ).…”

Qau-Net: Quartet Attention U-Net for Liver and Liver-Tumor Segmentation

“…To validate the superiority of the proposed QAU-Net, six state-of-the-art networks used for liver and liver-tumor segmentation are considered as comparative approaches. These networks can be grouped into two categories: 2D networks, 3D networks, where 2D networks include U-Net [3], U-Net++ [10], CE-Net [8] and MA-Net [14], 3D networks include 3D U-Net [15] and V-Net [16]. Table 2 and Table 3 present the segmentation performance on the test set using U-Net [3], CE-Net [8], MA-Net [14], 3D U-Net [15], V-Net [16], and proposed QAU-Net.…”

“…These networks can be grouped into two categories: 2D networks, 3D networks, where 2D networks include U-Net [3], U-Net++ [10], CE-Net [8] and MA-Net [14], 3D networks include 3D U-Net [15] and V-Net [16]. Table 2 and Table 3 present the segmentation performance on the test set using U-Net [3], CE-Net [8], MA-Net [14], 3D U-Net [15], V-Net [16], and proposed QAU-Net. It is clear that our QAU-Net achieves mean DICE of 96.13% and 85.90%, mean VOE of 8.52% and 24.13%, mean RVD of 1.85% and 0.82%, mean ASSD of 2.03 mm and 18.73 mm, and mean RMSD of 52.60 mm and 63.12 mm for liver and liver-tumor segmentation, respectively.…”

“…Chen et al [12] proposed FED-Net for automatic liver lesion segmentation that uses the SE block [13] to achieve the feature channel attention. Recently, Fan et al [14] proposed MA-Net that designed two blocks: Position-wise Attention Block (PAB) and Multi-scale Fusion Attention Block (MFAB) to capture the spatial dependencies and channel dependencies in a global view. Although these improved networks show better performance than U-Net on medical image segmentation, they can not capture the temporal information of volumetic data due to the employment of 2D convolution kernels, which may degrade the performance in volumetric segmentations.…”

AWEU-Net: An Attention-Aware Weight Excitation U-Net for Lung Nodule Segmentation

“…For the quantitative study, we used ACC, SEN, and SPE for pixel-level and DSC and IoU for object-level performance, respectively, as shown in Table 3. We compared the AWEU-Net to six different lung nodule segmentation models considering both datasets: PSPNet [40], MANet [41], PAN [42], FPN [43], DeeplabV3 [44], and U-Net [21,22]. As shown in Table 3, the integration of both PAWE and CAWE with the U-Net outperformed the segmentation results of the baseline model (U-Net).…”