Taichiro Shiodera1, Takamasa Sugiura1, Yuko Hara1, Yasunori Taguchi1, Tomoyuki Takeguchi1, Masao Yui2, Naotaka Sakashita2, Yasutaka Fushimi3, Takuya Hinoda3, Tomohisa Okada3, Aki Kido3, and Kaori Togashi3
1Toshiba Corporation, Kawasaki, Japan, 2Toshiba Medical Systems Corporation, Otawara, Japan, 3Kyoto University Graduate School of Medicine, Kyoto, Japan
We propose a background phase removal method for quantitative
susceptibility mapping using adaptive kernels depending on brain region.
Conventional methods use distance adaptive kernel spherical mean values
(SMV) to estimate background phase. However, artifacts occur where kernel sizes
are not optimal for certain brain regions. Here, we adapt SMV kernel sizes
depending on brain regions which are automatically detected by machine
learning methods. The proposed method eliminates tissue phase artifacts near air-tissue
interfaces in more central areas such as the sinus. The proposed method also eliminates streak
artifacts in susceptibility images.