[4C3-J-13] AI application: diagnosis

[4C3-J-13-03] Attenuation Correction of Brain SPECT using U-Net

〇Ryuhei Yamato¹, Taisuke Murata², Ryuna Kuryosawa², Joji Ohta², Takuro Horikoshi³, Hajime Yokota³, Yasukuni Mori⁴, Hiroki Suyari⁴ (1. Faculty of Engineering, Chiba University, 2. Dept. Radiology, Chiba University Hospital, 3. Dept. Radiology, Chiba University Hospital., 4. Graduate School of Engineering, Chiba University)

Keywords:SPECT, U-Net, Deep learning

SPECT is known to be one of nuclear medicine examinations. The attenuation problem in SPECT, the loss of detection of true coincidence events, increases image noises. The attenuation correction using CT is highly effective, but radiation exposure to patient cannot be avoided. In this paper, we propose a method to reproduce the attenuation correction using U-Net only instead of CT. For this purpose, we prepare one pair of SPECT images per one patient, uncorrected SPECT image and corrected SPECT image using CTAC. In our proposed method, the former image is given as input and the latter teacher image for the machine leaning. Our method successfully obtains the attenuation correction in SPECT image as almost same as CTAC by using machine learning only.