ASD-Net: a novel U-Net based asymmetric spatial-channel convolution network for precise kidney and kidney tumor image...

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computational Pathology

Background:

• Manual segmentation of kidneys and tumors in CT images is time-consuming and inconsistent.
• Deep learning offers potential for automated and accurate medical image segmentation.

Purpose of the Study:

• To develop an advanced deep learning network (ASD-Net) for precise kidney and kidney tumor segmentation.
• To overcome limitations of existing methods in segmenting complex and small kidney tumors.

Main Methods:

• Proposed ASD-Net incorporates novel Adaptive Spatial-channel Convolution Optimization (ASCO) and Dense Dilated Enhancement Convolution (DDEC) blocks.
• Integration of Atrous Spatial Pyramid Pooling (ASPP) and concurrent spatial and channel squeeze & excitation (scSE) attention mechanism.
• Enhanced U-Net architecture with additional encoding layers and a combined Binary Cross Entropy (BCE)-Dice loss function.

Main Results:

• ASD-Net demonstrated superior performance across multiple evaluation metrics on the KiTS19 dataset compared to existing segmentation networks.
• The model achieved high accuracy in kidney and kidney tumor segmentation tasks.
• ASD-Net secured second place in kidney tumor segmentation recall, closely following Attention-UNet.

Conclusions:

• The proposed ASD-Net network offers a robust and accurate solution for kidney and kidney tumor segmentation in medical imaging.
• The novel architectural components and loss function contribute to improved segmentation accuracy and detail restoration.
• ASD-Net shows significant promise for enhancing early tumor detection and improving patient outcomes.