@techreport{oai:ipsj.ixsq.nii.ac.jp:00191349, author = {北崎, 自然 and 川喜田, 雅則 and 實松, 豊 and 久原, 重英 and 竹内, 純一 and Shizen, Kitazaki and Masanori, Kawakita and Yutaka, Jitsumatsu and Shigehide, Kuhara and Jun'ichi, Takeuchi}, issue = {4}, month = {Sep}, note = {磁気共鳴画像法 (Magnetic Resonance Imaging ; MRI) は生体内の情報を得る有力な手段のひとつである.しかし,脳全体の 3 次元画像の取得には現在の一般的な MRI 装置で 30 分から 40 分程度必要であり,患者の負担軽減のため,撮影時間のさらなる短縮が求められている.この 10 年間,情報処理による MRI の高速化と高精細化に貢献してきたのが圧縮センシング (Compressed Sensing ; CS) であった. しかし,CS のアルゴリズムは観測数の 3 乗オーダーの計算量を必要とし,観測数の多いデータでは削減した検査時間以上の再構成時間を必要とするという欠点がある.本研究では,この問題に対し,深層ニューラルネットワークを利用した超解像により再構成時間の短縮を目指す., Magnetic Resonance Imaging (MRI) is one of the powerful techniques to acquire in vivo information. However, to obtain a three dimensional image of the whole brain, it takes thirty minutes to forty minites by using a current standard MRI scanner. Thus, to mitigate the inconvenience of the patient, further reduction of imaging time is required. For the past 10 years. Compressed Sensing (CS) has contributed to acceleration and high definition of MRI by means of information processing. However, CS has a disadvantage that the computatoinlal complexity of CS algorithms is the order of cubic of the number of samples. Thus, in the case of data with a large number of observations, it requires a reconstruction time longer than the reduced inspection time. In this research, we aim to reduce the computational time to reconstruct the MR image by image super-resolution using deep neural networks.}, title = {深層学習超解像を用いたMRI再構成の検討}, year = {2018} }