The　Deep　Learning　(DL)　method,　especially　the　Convolutional　Neural　Network　(CNN)　based　method　have　been　applied　in　various　fields　of　medical　image　processing　and　proven　to　be　powerful　for　image　classification,　denoising　task.　However,　there　are　not　any　end-to-end　DL　models　for　both　denoising　and　classification　tasks　in　existing　methods,　which　would　render　a　large　number　of　parameters　for　models　to　deal　with　these　tasks.　In　this　work,　we　proposed　an　efficient　auto-encoder　model　based　on　CNN　for　classification　and　denoising　tasks　for　brain　medical　images　(i.e.　Magnetic　Resonance　Imaging),　in　addition,　we　proposed　a　connection　between　encoding　and　decoding　units　in　our　auto-encoder　model　and　evaluated　the　contribution　of　it　to　denoising　task.　Our　method　achieved　99.7%　in　accuracy　of　classification　the　brain　medical　images　on　the　Brain　Tumor　dataset　from　Kaggle,　which　has　3,762　brain　medical　images,　in　addition,　the　denoising　performance　of　it　was　impressive,　as　the　lowest　average　Mean　Square　Error　(MSE)　of　the　dataset　was　0.00503　while　the　model　achieved　best　classification　performance.　©　2022　IEEE.