A Study Of The Feasibility Of Osteoporosis Evaluation Based On MRI And Texture Analysis

Part Ⅰ: Texture Characteristics Research on the Magnetic Resonance of Proximal Humerus Bone MarrowPurpose:To study the magnetic resonance signal evolution of proximal humerus bone marrow on the basis of texture analysis technique and to explore a new method to evaluate the rule of human bone marrow evolution.Materials and Methods:The imaging data of 232 patients who underwent magnetic resonance examination on shoulders were collected and the proximal humerus bone marrow signal was observed and divided into four types according to the bone marrow signal: homogeneous red bone marrow type(I),map type(Ⅱ),spot type(Ⅲ)and homogeneous yellow bone marrow type(IV).According to the median age,they were divided into high age group(>35 years old)and low age group(≤35 years old).And according to gender,they were divided into the male group and the female group.The software(ITK-SNAP)was adopted to sketch the proximal humeral bone marrow area as region of interest.The texture analysis software was adopted to make the analysis and the texture parameters with significantly statistical differences were obtained.Result:Among the 232 cases of bone marrow signals,61 cases(26.3%)were type I,49 cases(21.1%)were type Ⅱ,38 cases(16.4%)were type Ⅲ and 84 cases(36.2%)were type IV from visual classification.One texture parameter with significantly statistical difference was obtained in age groups,that is,Min Intensity.Six texture parameters with significantly statistical difference were obtained in gender groups,namely,Cluster Prominence_All Direction_offset4_SD,Cluster Shade_All Direction_offset7_SD,Inertia_All Direction_offset7_SD,Inertia_angle90_offset1,Inverse Difference Moment_All Direction_offset1,Short Run High Grey Level Emphasis_All Direction_offset7_SD.Conclusion:Quantitative analysis of proximal humerus bone marrow transformation by texture parameters can provide characteristic information other than conventional MR images and provide new quantitative indicators for clinical evaluation of humerus bone marrow lesions.Part Ⅱ: Feasibility of Opportunistic Osteoporosis Screening in Routine Spine MR Using RadiomicsPurpose: The purpose of our study is using texture analysis technique to screen opportunistic osteoporosis in routine lumbar MRI examination.Methods: The texture of lumbar vertebral were extracted from the sagittal position of MRI T1 WI and T2 WI.Using ANOVA and Mann-Whitney test,Spearman correlation analysis and LASSO dimension reduction method to carry out texture feature selection.According to the selected texture features,classifiers based on Logistic Regression was established in training set data.Finally,the model I was validated in testing set for estimate value of diagnosing osteoporosis.In addition,classifiers model Ⅱ based on radiomics combined with clinical risk factors and LASSO features was constructed in the training set and validation set.Results: A total of 136 subjects was included in,84 subjects were diagnosed as normal,52 subjects was diagnosed as osteoporosis,136 fourth lumbar vertebrae were included.770 texture features were extracted from vertebral body in T1 WI and T2 WI,After LASSO dimensionality reduction,8 texture parameters was selected.The classifier based on Logistic regression method shows performance.The AUC of training and testing set is 0.84 and 0.78.The accuracy of the model I is 0.71.The radiomics features combined with clinical risk factors and LASSO features showed a good predictive performance in the training and testing set is 0.87 and 0.86.The accuracy of the model Ⅱ is 0.81.Conclusion: Our study indicated that radiomics based on MRI could provide a good value to find osteoporosis in clinical practice.