Multi-sequence　cardiac　magnetic　resonance　(MS-CMR)　images　are　capable　of　providing　myocardial　pathology　information　for　patients　with　myocardial　infarction.　Precise　myocardial　structure　and　pathology　segmentation　hold　significant　importance　for　subsequent　diagnosis　and　treatment.　Nevertheless,　traditional　manual　myocardial　structure　and　pathology　segmentation　is　not　only　time-consuming　and　labor-intensive　but　also　has　a　low　accuracy　rate,　particularly　when　identifying　pathology　such　as　scars　and　edema　that　are　small　in　volume　and　have　low　contrast　with　the　surrounding　tissues,　it　becomes　even　more　challenging.　To　address　this　issue,　this　paper　proposes　an　improved　nn-UNet　for　fully　automatic　segmentation　of　myocardial　pathologies.　In　this　network,　based　on　nn-Unet,　we　use　multi-modal　data　as　input　to　make　up　for　the　lack　of　information　in　a　single　mode.　For　multi-modal　data,　we　utilize　cross　normalization　to　improve　the　generalization　performance.　Meanwhile,　multi-scale　attention　modules　are　integrated　to　process　features　at　different　resolutions,　thus　improving　the　feature　representation　capability　of　neural　networks.　Through　feature　fusion　and　attention　weighting,　the　model　can　better　capture　the　global　and　local　information　of　myocardial　pathologies,　and　achieve　more　accurate　segmentation　of　myocardial　pathologies.　To　verify　the　effectiveness　of　the　proposed　method,　we　conducted　an　evaluation　using　five-fold　cross-validation　in　the　dataset　of　the　MyoPS++　challenge.　©　The　Author(s),　under　exclusive　license　to　Springer　Nature　Switzerland　AG　2025.