Automatic Segmentation of the Prostate on 3D CT Images by Using Multiple Deep Learning Networks

Automatic segmentation of the prostate on CT images has many applications in prostate cancer diagnosis and therapy. However, prostate segmentation from CT images is a very challenging task due to the low contrast of soft tissue and the large variations of the prostate shape. In this paper, we propose an automatic segmentation method by using a two dimension (2-D) and a three dimension (3-D) convolutional neural network (CNN). First, instead of segmenting the whole image, we extract the volumes of interest (VOI) accurately to remove irrelevant regions by using a CNN based VOI extraction method (CBVEM). Then, we use the 3-D CNN to learn the holistic three-dimension deep features for distinguishing the prostate voxels from the non-prostate voxels in order to obtain the segmentation results. Deep learning networks can automatically learn the deep features based on the data, which are different from the handcrafted features. The proposed method has been evaluated on a dataset of 150 prostate CT images. Experimental results show that our method achieved a Dice similarity coefficient of 89.74% compared to the manual segmentation. Our deep learning based method is faster and returns similar results compared to those atlas-based, deformable model-based and feature-based classification methods. Due to the CBVEM and 3-D CNN, our method also achieves better performances on the same data in a minimum processing time compared to other deep CNN based methods. This proposed method for automatic segmentation of the prostate on 3-D CT images can have a variety of clinical applications.