Multimodal Clinical-radiomics Based On ¹⁸F-FDG Hybrid PET/MRI For Distinguishing Between Parkinson's Disease And Multiple System Atrophy

Objective: To construct multi-sequence radiomics models and select the optimal multimodal radiomics models step-by-step by using hybrid 18F-FDG PET/MRI for distinguishing between Parkinson's disease(PD)and multiple system atrophy(MSA).Methods: Ninety patients(60 with PD and 30 with MSA)were randomized to training and validation sets in a 7:3 ratio.All patients underwent 18F-Fluorodeoxyglucose(18F-FDG)PET/MRI to simultaneously obtain metabolic images(18F-FDG),structural MRI images(T1-weighted imaging [T1WI],T2-weighted imaging [T2WI] and T2-weighted fluid-attenuated inversion recovery [T2/flair])and functional MRI images(susceptibility-weighted imaging [SWI] and apparent diffusion coefficient).By using PET and five MRI sequences,we extracted 1172 radiomics features from the putamina and caudate nuclei.The radiomics signatures were constructed with the least absolute shrinkage and selection operator algorithm in the training set,with progressive optimization through random combinations of different sequences:(1)single-sequence radiomics signatures;(2)double-sequence radiomics signatures: 18F-FDG plus structural MRI(s MRI)sequences(18F-FDG + T1 WI,18F-FDG + T2 WI,18F-FDG + T2/FLAIR),or 18F-FDG plus functional MRI(f MRI)sequences(18F-FDG + SWI,18F-FDG + DWI);(3)optimal multimodal radiomics signatures: 18F-FDG + the best s MRI + the best f MRI,according to the results from the first and second steps.The diagnostic performance of the models was assessed by receiver operating characteristic(ROC),and the difference between every model was calculated by De Long test.Results: All the radiomics signatures showed favorable diagnostic efficacy with areas under curves(AUCs)> 0.8 in training and validation sets.Among the single-sequence models,the radiomics signature based on 18F-FDG(Radscore FDG)showed the best diagnostic efficiency,with the area under curves(AUCs)of the training and validation sets of 0.900 and 0.883,respectively.In the analysis of double-sequence models,Radscore FDG_T1WI had the highest diagnostic performance(AUCs of training and validation sets were 0.958 and 0.932,respectively)in the combinations of PET and s MRI;and Radscore FDG_SWI was the best(AUCs of training and validation sets were 0.951 and 0.951,respectively)in the combinations of PET and f MRI.Comparing the double-sequence models with the single-sequence models,the differences were statistically significant by using the De Long test(P < 0.05).The optimal radiomics signature was constructed by combined with 18F-FDG,T1WI(the best s MRI)and SWI(the best f MRI),and the AUCs were 0.971 and 0.957 in training and validation sets,respectively.De Long test showed the differences between Radscore FDG_T1WI_SWI with doublesequence models were statistically significant(P < 0.05).Conclusions: All the radiomics models achieve promising diagnostic efficacy for distinguishing between PD and MSA.The Radscore FDG_T1WI_SWI,with metabolic,structural,and functional information provided by hybrid 18F-FDG PET/MRI,is the optimal multimodal radiomics signature.Objective: To construct multivariate clinical-radiomics models using hybrid 18F-FDG PET/MRI for distinguishing between Parkinson's disease(PD)and multiple system atrophy(MSA),based on the optimal multimodal radiomics model(Radscore FDG_T1WI_SWI)in Part 1.Methods: Ninety patients(60 with PD and 30 with MSA)were the same as Part 1,and were randomized to training and validation sets in a 7:3 ratio.The information of clinical characteristics and SUV values were collected from all patients.The clinical characteristics included age,sex,disease duration(DD),age at onset,hypermyotonia,asymmetric symptoms at onset(ASO),bradykinesia,limbs tremor,dysarthria,and autonomic failure(AF).The SUV values included SUVmax,SUVmean,and SUVmin.The radiomics signature scores(Radscores)of all patients were calculated with the 18F-FDG PET/MRI images,based on the formula of the optimal multimodal radiomics model(Radscore FDG_T1WI_SWI)in Part 1.Multivariable logistic regression analysis was used to develop a clinical-radiomics model,combining the optimal multi-sequence radiomics signature with clinical characteristics and SUV values.The diagnostic performance of the models was assessed by receiver operating characteristic(ROC),calibration curves,and decision curve analysis(DCA).Finally,we recruited 6 patients(3 with PD,3 with MSA)as separate validation set to evaluate the accuracy and potential generalizability of the model.Results: There were no significant statistical differences between PD and MSA in most of clinical features(P > 0.05).The SUVmax and three clinical characteristics,including ASO,dysarthria,and AF,showed significant statistical differences in the training group(P < 0.05);but only dysarthria was retained in the validation group.The parameters of clinical-radiomics integrated model included three clinical features(DD,dysarthria,AF),SUVmax,and the Radscore.The clinical-radiomics model showed perfect diagnostic performance,with the AUCs of training and validation sets of 0.993 and 0.994,respectively.DCA indicated the highest clinical benefit of the clinical-radiomics integrated model,better than purely radiomics(Radscore FDG_T1WI_SWI),while the purely clinical model was the lowest.The results of the prospective separate validation set showed that the high accuracy of differentiating PD and MSA was improved in both the radiomics and clinical-radiomics models.Conclusion(s): The clinical-radiomics integrated model based on the hybrid 18F-FDG PET/MRI,combining clinical features and multimodal imaging information(metabolic,structural,and functional information),may achieve better diagnostic efficacy and clinical benefit for distinguishing between PD and MSA,and have some clinical application value.