Segmentation　of　gliomas　is　a　crucial　step　in　brain　tumor　surgical　planning,　and　it　serves　as　the　foundation　for　further　diagnosis　of　brain　tumors.　Tumor　borders　are　usually　unclear,　and　a　significant　amount　of　heterogeneity　in　the　structure,　causing　brain　tumor　segmentation　a　tough　task.　However,　for　tumor　segmentation,　approaches　based　on　deep　learning　have　shown　promising　results.　This　study　develops　a　multi-module　U-Net　system　that　utilizes　multiple　U-Net　modules　to　collect　spatial　detail　at　varying　resolutions.　We　use　various　up-inception　and　down-inception　modules　to　extract　and　exploit　enough　features.　Experimental　results　show　that　the　dice　scores　of　0.95,　0.90,　0.84,　and　0.91,　0.84,　0.77　were　achieved　for　the　whole　tumor,　core　tumor,　and　enhancing　tumor,　using　the　BraTS　2018　and　local　private　dataset,　respectively.　When　compared　to　cutting-edge　methods,　this　study　achieves　competitive　segmentation　results.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.