Hsian-Min Chen1, Siwa Chan2,
Jyh-Wen Chai2, Clayton Chi-Chang Chen2, San-Kan Lee2,
Chein-I Chang3, Min-Ying Su4, Orhan Nalcioglu4,
Jeon-Hor Chen4,5
1Department of Biomedical
Engineering, HungKuang University, Taichung, Taiwan; 2Department
of Radiology, Taichung Veterans General Hospital, Taichung, Taiwan; 3Department
of Computer Science & Electrical Engineering, University of Maryland,
Baltimore, United States; 4Center for Functional Onco-Imaging,
University of California Irvine, California, United States; 5Department
of Radiology, China Medical University Hospital, Taichung, Taiwan
A supervised multispectral analysis of breast density in MRI using ICA+SVM technique was developed. With this approach, two sets of images (T1WI and T2WI in this study) were needed for the analysis. ICA was used to enhance the image contrast and used as a preprocessing method to separate different tissue. SVM was used as a binary classifier to maximize the margin between two classes of data samples. In this study we have shown that the intra- and inter-operator measurement variation is very small. The high consistency of this method can be potentially applied for evaluation of small breast density change in longitudinal follow-up study.