Ultrasonography　is　a　valuable　diagnosis　method　for　thyroid　nodules.　Automatically　discriminating　benign　and　malignant　nodules　in　the　ultrasound　images　can　provide　aided　diagnosis　suggestions,　or　increase　the　diagnosis　accuracy　when　lack　of　experts.　The　core　problem　in　this　issue　is　how　to　capture　appropriate　features　for　this　specific　task.　Here,　we　propose　a　feature　extraction　method　for　ultrasound　images　based　on　the　convolution　neural　networks　(CNNs),　try　to　introduce　more　meaningful　semantic　features　to　the　classification.　Firstly,　a　CNN　model　trained　with　a　massive　natural　dataset　is　transferred　to　the　ultrasound　image　domain,　to　generate　semantic　deep　features　and　handle　the　small　sample　problem.　Then,　we　combine　those　deep　features　with　conventional　features　such　as　Histogram　of　Oriented　Gradient　(HOG)　and　Local　Binary　Patterns　(LBP)　together,　to　form　a　hybrid　feature　space.　Finally,　a　positive-sample-first　majority　voting　and　a　feature-selected　based　strategy　are　employed　for　the　hybrid　classification.　Experimental　results　on　1037　images　show　that　the　accuracy　of　our　proposed　method　is　0.931,　which　outperformed　other　relative　methods　by　over　10%.