As　domain　shift　hinders　deep　learning　models＇　performance　on　medical　image　analysis,　Domain　Generalization　(DG)　has　been　proposed　to　enhance　the　generalization　capability.　Traditional　multi-source　domain　generalization　methods　require　the　training　data　from　multiple　source　domains,　which　is　often　time-consuming　and　laborious　to　collect.　So,　considering　that　CT　and　MR　are　two　indispensable　medical　image　domains,　we　propose　a　Enhancing　Single-Source　Domain　Generalization　from　CT　to　MR　by　data　augmentation　based　on　grayscale　distribution　remapping　to　perform　tasks　of　medical　image　classification　and　segmentation.　First,　we　put　forward　Distribution　Remapping　Module　(DRM)　to　realize　effective　data　augmentation　through　generating　various　grayscale　distributions　based　on　CT　images,　thus　simulating　various　grayscale　distributions　of　medical　images.　Then,　each　original　medical　image　and　its　corresponding　augmented　image　form　a　positive　pair.　The　contrastive　loss　calculated　by　these　pairs,　together　with　the　task　loss　for　medical　image　classification　or　segmentation　task,　is　utilized　to　optimize　the　model.　Compared　to　existing　state-of-the-art　methods,　experiments　on　three　benchmark　datasels,　i.e.　TotalSegmentator,　CHAOS,　and　AMOS,　achieve　best　performance　in　both　organ　classification　and　segmentation　tasks.