Feasibility of Magnetic Resonance Imaging Features in Identifying Patients with Spastic Cerebral Palsy for GMFCS Classification
Objective GMFCS grading is the main grading method for evaluating the level of motor function in spastic cerebral palsy and is widely used in clinical decision making.However,how to improve the precision of GMFCS grading judgment is an important issue.In this study,we aimed to construct an imaging histology model using calf triceps MRI to achieve differentiation between GMFCS grades Ⅰ-Ⅴin patients with SCP.Methods With the approval of the Ethics Committee,16 patients with GMFCS grade Ⅳ-Ⅴ SCP(10 males and 6 females)and 40 patients with GMFCS grade Ⅰ-Ⅲ SCP(28 males and 12 females)were collected in this study.T2-weighted imaging using calf MRI was utilized for analysis.After manually segmenting the calf triceps,the image features were screened using methods such as LASSO regression,and four methods,namely,the linear model LR,KNN,the tree model XGBoost,and the deep learning model MLP,were used to model and evaluate the model performance.Results Log-sigma-20mm-3D firstorder-maximum,Log-sigma-20mm 3D glcm-ldn,and Wavelet-LLH-glszm-SizeZoneNonUniformity are the core features that can differentiate the GMFCS classification.When evaluating the model performance,it performs exceptionally well in the XGBoost model,with an AUC of 0.981 in the training dataset,but drops to 0.729 in the test dataset.it has high sensitivity(0.958)and specificity(0.923)in the test dataset.When assessing model fit using the Hosmer-Lemeshow test,all models except the KNN model exhibited p-values greater than 0.05 in both the training and test cohorts,demonstrating the reliability and validity of the models in predicting outcomes.Each model was fully evaluated using Decision Curve Analysis(DCA),again demonstrating significant advantages.Conclusion Constructing a diagnostic model for GMFCS grading by calf triceps MRI imaging histology is a promising approach with implications in improving the precision of GMFCS grading judgments.
NETL
NSTL
万方数据
维普
