With　the　explosive　growth　of　lung　diseases　in　patients,　automatically　detecting　diseases　and　obtaining　accurate　diagnosis　through　the　X-ray　medical　images　become　the　new　research　focus　in　the　field　of　computer　science　and　artificial　intelligence　to　save　the　significant　cost　of　manual　labeling　and　classifying.　However,　the　quality　of　common　radiograph　is　not　satisfied　for　the　most　tasks,　and　traditional　methods　are　deficient　to　deal　with　the　massive　images.　Therefore,　we　present　a　feature　fusion　convolutional　neural　network　(CNN)　model　to　detect　pneumothorax　from　chest　X-ray　images.　Firstly,　the　preprocessed　image　samples　are　enhanced　by　two　methods.　Then,　a　feature　fusion　CNN　model　is　introduced　to　combine　the　Gabor　features　with　the　enhanced　information　extracted　from　the　images　and　implement　the　final　classification.　Comprehensive　qualitative　and　quantitative　experiments　demonstrate　that　our　proposed　model　achieve　better　results　in　multi-angle　views.