Comparisons of fat quantification methods based on MRI segmentation

The improvement of the quality of life brings people not only a lot of convenience, but also some bad habits which contribute to some fatal cardiovascular diseases. And it is also proved that the high fat content of tissues has a close relationship with some undesirable diseases, such as the Diabetes, Obesity, Hypertension and so forth. Current approaches to measure body fat content are limited and lack accuracy with traditional methods, including Skin Fold method, Fat-Soluble Gases measurement, Underwater Measurement and Electrical Impedance. Since adipose can be highlighted in MRI due to its imaging characteristic, MRI began to be widely applied to fat quantification. However, manual analysis of MRI data is time-consuming and likely to produce subjective errors. Therefore, research on automatic reorganization of fat distribution attracts many efforts. Precision of image segmentation determines the accuracy of fat calculation. Due to the several challenges: the inhomogeneous image degenerates the image quality, the poor histogram separation of different tissues and the shape differences between subjects, and it is hard to get an accuracy result of segmentation. There are many available algorithms for image segmentation. However, few objective evaluations exist of these segmentation algorithms. To fill this gap, this paper presents an evaluation of the methods utilized broadly in the relevant fields, including Watershed Segmentation, Region Growing Segmentation and Threshold Segmentation applied to 33 MRI data analysis. The evaluation of these methods offers reference for its application in MRI fat segmentation.