Biomedical Image Analysis with Deep Neural Networks
Project Description:
Previously, tremendous works have done for liver segmentation and tumor detection. However, most procedures require manual interactions. As the deep learning has significant performance in image classification and segmentation. Several researchers have tried to apply this hot technique to the medical image area to help doctors do a fast and coarse diagnosis. We would like to find a new deep learning architecture to show better performance compared with other architectures so far.
Project Duration:
11/2017 to 11/2018
Project Significance:
Current state-of-art achievements for this liver tumor segmentation would only achieve around 65~70% dice similarity coefficients. Although this result already beats previous traditional methodologies, it is still not quite satisfied to doctors. New architectures would hopefully give better performance than current works.
Remarks:
Deep learning is a highly data-hungry algorithm. It requires a huge amount of data feed into the architecture, which needs large memory to store the data. Besides, the architecture has tremendous layers and asking for high-performance computing machines, such as GPU.