Among　many　molecular　imaging　modalities,　bioluminescence　tomography　(BLT)　stands　out　as　an　effective　approach　for　in　vivo　imaging　because　of　its　noninvasive　molecular　and　cellular　level　detection　ability,　high　sensitivity　and　low　cost　in　comparison　with　other　imaging　technologies.　However,　there　exists　the　case　that　large　scale　problem　with　large　number　of　points　and　elements　in　the　structure　of　mesh　standing　for　the　small　animal　or　phantom.　And　the　large　scale　problem＇s　system　matrix　generated　by　the　diffuse　approximation　(DA)　model　using　finite　element　method　(FEM)　is　large.　So　there　wouldn＇t　be　enough　random　access　memory　(RAM)　for　the　program　and　the　related　inverse　problem　couldn＇t　be　solved.　Considering　the　sparse　property　of　the　BLT　system　matrix,　we＇ve　developed　a　new　sparse　matrix　(ZSM)　to　overcome　the　problem.　And　the　related　algorithms　have　all　been　speeded　up　by　multi-thread　technologies.　Then　the　inverse　problem　is　solved　by　Tikhonov　regularization　method　in　adaptive　finite　element　(AFE)　framework.　Finally,　the　performance　of　this　method　is　tested　on　a　heterogeneous　phantom　and　the　boundary　data　is　obtained　through　Monte　Carlo　simulation.　During　the　process　of　solving　the　forward　model,　the　ZSM　can　save　more　processing　time　and　memory　space　than　the　usual　way,　such　as　those　not　using　sparse　matrix　and　those　using　Triples　or　Cross　Linked　sparse　matrix.　Numerical　experiments　have　shown　when　more　CPU　cores　are　used,　the　processing　speed　is　increased.　By　incorporating　ZSM,　BLT　can　be　applied　to　large　scale　problems　with　large　system　matrix.