Automatic　airway　segmentation　is　essential　for　quantitative　measurement　of　lung　diseases.　However,　most　researches　mainly　suffer　from　two　challenges:　the　spatial　distribution　and　various　branch　pattern　of　airway,　topology　preserving　of　the　segmentation.　We　propose　a　feature　aggregation　and　topology　embedding　network　(FATENet),　which　can　combinate　topology　information　and　attention　mechanism　to　mitigate　those　problems.　Topology　features　are　embedded　into　our　network　as　a　loss　function,　and　attention　mechanism　are　utilized　to　capture　global　dependencies.　Topology　features　are　embedded　into　our　network　as　a　loss　function,　and　attention　mechanism　are　utilized　to　capture　global　dependencies.　This　allows　the　network　to　generate　more　informative　output　and　improve　the　segmentation　performance.　Our　method　is　evaluated　on　25　cases　from　different　data　resources.　As　compared　to　some　existing　method　and　network,　we　achieve　the　Dice　similarity　coefficient　(DSC),　false　positive　rate　(FPR),　and　true　positive　rate　(TPR)　are　93.4%,　0.054%,　and　92.5%,　respectively.　The　result　shows　that　our　method　have　accurate　and　efficient　performance.　©　2021　IEEE.