Automatic Liver Segmentation of CT Volumes Using 2D Attention-UNet
References:Attention gated networks: Learning to leverage salient regions in medical images
- Automatic segmentation of the liver is an important step towards deriving quantitative biomarkers for accurate clinical diagnosis and computer-aided decision support systems.
- Used the 2D U-Net architecture integrating attention mechanism and dice loss for accurate liver segmentation.
- Understood and processed the image data using Numpy and Scikit-Image. Implemented Attention-UNet using PyTorch.
- Achieved Dice scores over 96% for the liver in LiTS challenge dataset with computation times below 100s per volume. Applied the technology to Nanjing Zhongda Hospital to help doctors in clinical research.