Cherenkov-excited　luminescence　scanned　imaging　(CELSI)　is　a　new　emerging　imaging　modality,　which　uses　linear　accelerator　(LINAC)　to　induce　Cherenkov　radiation,　and　then　secondary　excite　molecular　probes　to　produce　luminescence.　The　tomographic　distribution　of　the　molecular　probes　can　be　recovered　by　a　reconstruction　algorithm.　However,　the　reconstruction　images　usually　suffer　from　many　artifacts.　To　improve　the　image　quality　for　tomographic　reconstruction,　we　propose　a　reconstruction　method　based　on　learned　KSVD.　Numerical　simulation　experiments　reveal　that　the　proposed　algorithm　can　reduce　the　artifacts　in　the　reconstructed　image.　The　quantitative　results　show　that　the　structured　similarity　(SSIM)　is　improved　more　than　8.8%　compared　to　the　existing　algorithms.　In　addition,　our　results　also　demonstrate　that　the　proposed　algorithm　has　the　best　performance　under　different　noise　levels　(0.5%,　1%,　2%,　and　4%).