Objective To observe the value of MR 3D-T1WI and T2WI radiomics for diagnosing early and middle stage Parkinson disease(PD).Methods A total of 96 patients with early or middle stage PD(Hoehn-Yahr[H-Y]stage≤2.5)and 96 matched healthy adults were retrospectively collected and divided into training set(n=135,including 67 cases of PD and 68 healthy adults)and validation set(n=57,including 28 cases of PD and 29 healthy adults)at the ratio of 7∶3.The optimal radiomics features of left red nucleus(LRN),right red nucleus(RRN),left substantia nigra(LSN)and right substantia nigra(RSN)were extracted and screened from cranial 3D-T1WI and T2WI in training set.Then radiomics models of single MR sequence and combined MR sequences were constructed,respectively,the radiomics scores(Radscore)were obtained,the diagnostic efficacy of each model for diagnosing early and middle stage PD was validated using validation set,and the correlations of Radscore of each model and clinical scale scores of PD patients were analyzed.Results Based on LRN,RRN,LSN and RSN on 3D-T1WI and T2WI,15,14,11 and 14,and 15,12,14 and 12 optimal radiomics features were obtained,respectively.Then models of single sequence,including LRN3D-T1W,I RRN3D-T1W,I LSN3D-T1W,I RSN3D-T1W,I LRNT2W,I RRNT2W,I LSNT2WI and RSNT2W,I as well as models of combined sequences,including LRN3D-T1WI+T2WI,RRN3D-T1WI+T2WI,LSN3D-T1WI+T2WI and RSN3D-T1WI+T2WI were constructed.The AUC of models in training and validation sets based on 3D-T1WI were 0.75-0.86,of models based on T2WI in training and validation sets were 0.82-0.90,while of combined models were 0.85-0.93.The Radscore of LRN3D-T1WI model in PD patients was negatively correlated with Hamilton depression scale(HAMD)and Hamilton anxiety scale(HAMA)scores(rs=-0.255,-0.242,P=0.011,0.016),while of RSNT2WI model was negatively correlated with HAMD score(rs=-0.254,P=0.010).Conclusion 3D-T1WI and T2WI radiomics could be used to diagnose early and middle stage PD.

Objective To observe the value of MR 3D-T1WI and T2WI radiomics for diagnosing early and middle stage Parkinson disease(PD).Methods A total of 96 patients with early or middle stage PD(Hoehn-Yahr[H-Y]stage≤2.5)and 96 matched healthy adults were retrospectively collected and divided into training set(n=135,including 67 cases of PD and 68 healthy adults)and validation set(n=57,including 28 cases of PD and 29 healthy adults)at the ratio of 7∶3.The optimal radiomics features of left red nucleus(LRN),right red nucleus(RRN),left substantia nigra(LSN)and right substantia nigra(RSN)were extracted and screened from cranial 3D-T1WI and T2WI in training set.Then radiomics models of single MR sequence and combined MR sequences were constructed,respectively,the radiomics scores(Radscore)were obtained,the diagnostic efficacy of each model for diagnosing early and middle stage PD was validated using validation set,and the correlations of Radscore of each model and clinical scale scores of PD patients were analyzed.Results Based on LRN,RRN,LSN and RSN on 3D-T1WI and T2WI,15,14,11 and 14,and 15,12,14 and 12 optimal radiomics features were obtained,respectively.Then models of single sequence,including LRN3D-T1W,I RRN3D-T1W,I LSN3D-T1W,I RSN3D-T1W,I LRNT2W,I RRNT2W,I LSNT2WI and RSNT2W,I as well as models of combined sequences,including LRN3D-T1WI+T2WI,RRN3D-T1WI+T2WI,LSN3D-T1WI+T2WI and RSN3D-T1WI+T2WI were constructed.The AUC of models in training and validation sets based on 3D-T1WI were 0.75-0.86,of models based on T2WI in training and validation sets were 0.82-0.90,while of combined models were 0.85-0.93.The Radscore of LRN3D-T1WI model in PD patients was negatively correlated with Hamilton depression scale(HAMD)and Hamilton anxiety scale(HAMA)scores(rs=-0.255,-0.242,P=0.011,0.016),while of RSNT2WI model was negatively correlated with HAMD score(rs=-0.254,P=0.010).Conclusion 3D-T1WI and T2WI radiomics could be used to diagnose early and middle stage PD.