Vestibule　Segmentation　is　of　great　significance　for　the　clinical　diagnosis　of　congenital　ear　malformations　and　cochlear　implants.　However,　automated　segmentation　is　a　challenging　task　due　to　the　tiny　size,　blur　boundary,　and　drastic　changes　in　shape　and　size.　In　this　paper,　a　vestibule　segmentation　method　from　CT　images　has　been　proposed　specifically,　which　exploits　different　deep　feature　fusion　strategies,　including　convolutional　feature　fusion　for　different　receptive　fields,　channel　attention　based　feature　channel　fusion,　and　encoder-decoder　feature　fusion.　The　experimental　results　on　the　self-established　vestibule　segmentation　dataset　show　that,　compared　with　several　state-of-the-art　methods,　our　method　can　achieve　superior　segmentation　accuracy.