A 3D Fully Convolutional Network Based Semantic Segmentation for Ear Computed Tomography Images

Ear computed tomography (CT) has become an important means of diagnosing ear diseases, which provides doctors with a chance of observing the shape and components of the key anatomical structures of the auditory system. Therefore, it is helpful to diagnose the ear diseases early. However, the anatomical structures of the auditory system are characterized by complexity, sophisticated, and large individual differences, meanwhile, they are small and difficult to segment. Most of the existing medical image segmentation algorithms fail in segmenting the ear anatomical structures. To address the problem, a 3D fully convolutional network (3D- FCN) based semantic segmentation method is proposed for the key anatomical structures of ear CT Images. We evaluated our approach on the ear CT dataset. Compared to the 2D fully convolutional network (2D-FCN), the mean Dice-Serensen Coefficient (DSC) of our method is improved significantly in the task of segmentation for six key anatomical structures of the ear. The experimental results show that our method can effectively improve the segmentation accuracy of key anatomical structures of ear CT images.