Automatic background-foreground segmentation of organs on MRI images of the head-neck area

Accurate segmentation of organs on MRI sequences can be crucial in patient treatment. It supports the diagnosis of diseases and helps in treatment planning. For example, in oncology, analyzing the shape of the organs can give valuable information on the extent of the tumors, and in radiation therapy planning, the delineation of organs of risk is an indispensable first step of the treatment. In this paper, we give a probability atlas, and deep convolutional network-based method for the automatic segmentation of six organs (trachea, spinal cord, parotid glands, sternocleidomastoid muscle, arteria carotis commulus and vena jugularis interna) on multi-sequential MRI images of the head-neck area. The method was also evaluated on clinical data and found to give accurate results according to the Dice metric.