MyoPS　(Myocardial　Pathology　Segmentation)　is　used　for　auxiliary　diagnosis　of　myocardial　infarction　by　accurately　segmenting　myocardial　lesions　(such　as　scars　and　edema).　However,　CMR　images　are　complex,　manual　segmentation　is　time-consuming　and　relies　on　professional　knowledge,　and　there　are　differences　in　imaging　data　from　different　centers,　which　increases　the　difficulty　of　segmentation.　To　this　end,　this　study　developed　a　domain　generalization　module　that　flexibly　integrates　LGE,　T2-weighted,　and　Cine　sequences　to　improve　cross-center　and　multi-sequence　adaptability　and　robustness.　Our　method　combines　the　domain　generalization　module　with　the　nnUNet　segmentation　network,　and　reduces　the　differences　between　different　data　distributions　by　utilizing　the　domain　generalization　module　for　data　mixing　enhancement,　thereby　enhancing　the　model’s　generalization　ability　and　improving　segmentation　performance.　In　tests　conducted　on　the　data　set　of　the　MyoPS++　Challenge,　our　network　performed　well　in　segmenting　scars　and　edema.　Compared　with　the　native　segmentation　network,　it　has　a　greater　performance　improvement,　which　verifies　its　ability　to　handle　multi-center,　Effectiveness　in　multi-sequence　CMR　data.　©　The　Author(s),　under　exclusive　license　to　Springer　Nature　Switzerland　AG　2025.