Research And Application Of Machine Learning In Adipose Tissue Quantification Of MRI

With the improvement of living standard,health has become a more and more concerned issue.Therefore,accurate measurement of subcutaneous adipose tissue and visceral adipose tissue is of great significance for the study of many diseases.IDEAL-IQ technology in MRI is a safe and painless examination method mainly used for imaging the adipose tissue in the scanning area,and generates six sequences in one scan.After decades of development,computer aided diagnosis technology has become a comprehensive use of a variety of advanced technology of clinical diagnosis tool.Among the algorithms covered by computer aided diagnosis,machine learning is becoming the preferred method for medical image data processing,which can be applied to a variety of different radiology imaging tasks.The purpose of this paper is to explore the quantitative methods of subcutaneous adipose tissue and visceral adipose tissue and related segmentation based on IDEAL-IQ technology imaging and machine learning.In this paper,an automatic segmentation method based on 2D slice is designed.Firstly,the U-Net network is used to segment the inner and outer contours of subcutaneous adipose tissue and the peritoneal cavity in the water image;secondly,the contour of peritoneal cavity is mapped to the fat image of the same body region,and AFK-MC~2 clustering method is used to quantify the visceral adipose tissue,which has achieved good results.On the basis of segmentation results,this paper quantifies adipose related indicators.Among them,this paper proposes a semi-automatic subcutaneous adipose thickness measurement method based on the later midline,which corrects the incorrect position of the subject;then,this paper carries out a fully automatic measurement of abdominal circumference length.In order to add the information between slices into the deep learning model,a 3D U-Net based network is designed to segment and quantify the subcutaneous adipose tissue and visceral adipose tissue.Finally,two-dimensional and three-dimensional fully automatic subcutaneous adipose tissue and visceral adipose tissue segmentation model are embedded into the RIAS platform of College of Electronic Science and Engineering,Jilin University and Philips ISD platform,which not only realizes the application of automatic segmentation model in clinical computer aided diagnosis,but also enhances the functional diversity of two medical image research platforms.The experimental results show that the automatic quantitative method used in this paper has high accuracy and robustness,and has high reliability with the results of manual measurement.Firstly,the Dice coefficients of subcutaneous adipose tissue and visceral adipose tissue obtained from the two-dimensional model are 0.96 and0.97 respectively;secondly,the Dice coefficients of 3D U-Net segmentation of subcutaneous adipose tissue and visceral adipose tissue are 0.98 and 0.90 respectively.