Ankle　joint　injury　is　a　kind　of　common　and　frequently-occurring　clinical　musculoskeletal　injury　disease.　The　clinical　diagnosis　of　ankle　joint　injury　is　mostly　based　on　conventional　computed　tomography　(CT),　X-ray　and　doctor＇s　consultation.　However,　because　of　the　insufficient　understanding　and　cognition　of　ankle　joint　injury　by　some　physicians,　missed　diagnosis　and　misdiagnosis　are　often　caused.　Therefore,　the　correct　treatment　measures　and　treatment　time　are　always　missed,　making　the　acute　ankle　joint　injury　become　chronic　ankle　instability.　Surface　electromyography　(sEMG)　is　a　non-stationary　weak　bioelectrical　signal　that　is　superimposed　on　the　surface　of　the　skin　by　the　action　of　potential　sequence　generated　by　muscle　excitability　during　exercise.　SEMG　has　great　performance　on　the　recognition　of　human　movements　and　the　diagnosis　of　injuries　because　of　its　strong　muscle　specificity　and　differences　in　exercise　patterns.　In　this　paper,　based　on　the　difference　of　the　information　provided　by　sEMG　measured　by　patients　and　healthy　people,　an　intelligent　diagnosis　method　is　proposed　to　help　doctors　improve　the　accuracy　of　diagnosis.　This　intelligent　diagnostic　method　is　named　cerebellar　model　neural　network　(CMNN)　that　imitates　the　learning　mechanism　of　the　human　cerebellum.　The　trained　CMNN　can　be　seen　as　an　expert　who　is　good　at　ankle　diagnosis　and　can　help　doctors　make　a　better　diagnose.　This　study　reduces　the　number　of　iterations　and　speeds　up　the　diagnosis　by　changing　the　loss　function　of　CMNN　from　mean　square　error　(MSE)　to　cross　entropy.