Recent　years　CNN　(Convolutional　Neural　Network)　has　performed　well　in　image　processing,　including　image　retrieval.　However,　since　the　features　of　CNN　extraction　are　usually　high-dimensional,　and　in　the　massive　data　conditions,　it　is　a　rather　time-consuming　process　to　traverse　all　the　images　and　calculate　the　distance　between　the　feature　vectors　to　accurately　find　the　closest　Top　K　images.　The　proposed　paper　uses　an　effective　deep　learning　framework　in　which　Deep　Convolution　Network　is　combined　with　Hash　Coding　to　learn　rich　medical　image　representing　through　CNN.　First,　a　hash　layer　is　added　to　the　network　to　represent　the　image　information　as　binary　hashing　codes;　Simultaneously,　the　dimension　of　feature　vector　is　effectively　reduced　by　the　framework;　then,　In　order　to　improve　the　accuracy　of　image　retrieval,　rough　searching　and　fine　searching　are　combined.　The　experimental　results　show　that　our　method　is　optimal　than　several　hashing　algorithms　and　CNN　methods　on　the　TCIA-CT　database　and　VIA/I-ELCAP　database.　©　2017　IEEE.