Automatic segmentation of anatomical structures in three-dimensional CT images

What is it about?

Anatomical structures are important for diagnosis, radiotherapy, and surgery planning that are major tasks of healthcare. Understanding three-dimensional (3D) structures via the human observation in the 3D gray images is subjective, time-consuming and costly. The computer-based segmentation of human organs in 3D medical images was desired to do for the further applications but it is still unsolved now​. We solved this problem by using a modern deep learning approach.

Featured Image

Why is it important?

We propose a single network based on label-to-pixel deep learning to address the challenging issue of anatomical structure segmentation in 3D CT cases. The unique point of this work is the policy of deep learning of the different 2D sectional appearances of 3D anatomical structures for CT cases and the majority voting of the 3D segmentation results from multiple crossed 2D sections to achieve availability and reliability with better efficiency, generality, and flexibility than conventional segmentation methods, which must be guided by human expertise. This network was applied to segment 19 types of targets in 240 3D CT scans, demonstrating highly promising results. This work is the first to tackle anatomical segmentation (with a maximum of 19 targets) on scale-free CT scans (both 2D and 3D images) through a single deep neural network.

Perspectives