Magnetic　resonance　imaging　(MRI)　is　currently　the　main　non-invasive　method　for　detecting　focal　liver　lesions　(FLLs)　as　it　can　provide　rich　information　from　multiple　modals.　Although　deep　learning　has　made　significant　progress　in　medical　image　diagnosis,　medical　image　datasets　rarely　contain　large　scale　labelled　data　which　often　leads　to　overfitting　and　poor　model　generalization.　In　order　to　make　full　use　of　the　multimodal　MRI　under　few-shot　scenarios,　we　propose　an　Efficient　Multimodal-Contribution-Aware　N-pair　(EMCAN)　network,　which　constructs　a　lightweight　and　efficient　feature　extractor　to　enhance　representation　of　features.　To　improve　the　separability　of　the　features　of　this　network,　we　propose　the　multi-class　N-pair　loss.　Experimental　results　show　that　our　method　outperforms　conventional　deep　learning　models　in　terms　of　diagnostic　accuracy　and　provides　more　accurate　reference　for　clinical　diagnosis.　©　2023,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.