Research on dual encoding DAM-Unet liver tumor segmentation based on attention and multi-scale feature fusion

Yanmin Niu ^* Jianfeng Li ^*

• Chongqing Normal University, Chongqing, China

A dual encoding DAM-Unet liver tumor segmentation model based on attention and multi-scale feature fusion is proposed to address the issue of inaccurate image segmentation caused by inconsistent lesion size, unclear boundary lines, and overly complex network in medical images. In order to capture more detailed information, a layer by layer fusion residual convolution block (FRC) was innovatively proposed to replace traditional convolutional blocks for convolution of the right encoding in dual encoding, while effectively solving the problem of gradient vanishing or gradient explosion; Multi scale feature fusion provides more comprehensive and accurate contextual information, improving network segmentation performance; A Jump Connection Fusion (JCF) module was designed to reduce the interference of invalid features by applying weight ratios to important features through spatial and channel attention. The experimental results on the LiTS dataset show that the Dice coefficient, mean Intersection over Union, Precision, Recall, and Accuracy of the proposed algorithm are 89.70%, 89.14%, 91.87%, 94.97%, and 98.95%, respectively.