深度學習在近期被廣泛的運用在不同領域的研究之中，其中也不乏有應用在MRS的研究，如利用卷積神經網路(CNN)來對頻譜品質來做好壞分類，亦或是利用CNN模型來達到去除雜訊、移除基線與譜線寬度縮減等減緩缺陷來達成代謝物濃度相對定量的研究。為此，本研究更進一步探討利用CNN模型搭配water scaling 的方式來達成代謝物的絕對濃度定量，藉由模擬的人類大腦頻譜來訓練模型，以達成還原純代謝物的頻譜目的，再利用多元線性回歸來算出代謝物的濃度定量。在評估模擬頻譜的測試集資料上，我們採用平均絕對百分誤差(MAPEs)來衡量定量結果。接著更進一步驗證提出的方法在in vivo 資料上，挑選三個腦區以驗證在不同頻譜品質的效果，並以套裝軟體LCModel的定量結果做比較。結果顯示使用CNN模型搭配多元線性回歸所定量出的濃度大小，在tNAA, Cr, tCho, mI, Glu這幾種代謝物與LCModel定量出的結果有相似的範圍，並與過往研究有相符的結果。並在不同受測資料間有較小的變異數，皮爾森相關係數的統計結果亦有顯著的相關。

Recently, it has been shown that MRS can be analyzed by a convolutional neural network (CNN) for spectral quality sorting or metabolite quantified in a relative way. Here, we propose a strategy to scale in vivo MRS data by water signal to achieve absolute metabolite quantification. The CNN model was trained by simulated human brain spectra and quantified the metabolite concentration by multiple linear regression. The result are evaluated by the mean‐absolute‐percent‐errors (MAPEs) on simulated test data set. And we further validated the proposed method on in vivo data with 3 regions in brain as different data quality. The result show that the concentrations of tNAA, Cr, tCho, mI, Glu are in similar range with statistically significant Pearson’s correlation coefficient. The between subject variations of these metabolite reported by CNN are in compatible range with the previous studies.

[1]. Govindaraju V, Young K, Maudsley AA. Proton NMR chemical shifts and coupling constants for brain metabolites. NMR Biomed 2000;13(3):129-53.
[2]. Xue Q, Shen Z-w, Wang H, Chen Y, Wu R. 1H-MR spectroscopy quantification using LCModel with the simulated basis-set created by software Vespa. Proceedings - 2011 4th International Conference on Biomedical Engineering and Informatics, BMEI 2011 2011;1:572-5.
[3]. Provencher SW. Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn Reson Med 1993;30(6):672-9.
[4]. Minati L, Aquino D, Bruzzone MG, Erbetta A. Quantitation of normal metabolite concentrations in six brain regions by in-vivoH-MR spectroscopy. J Med Phys 2010;35(3):154-63.
[5]. Barker PB, Soher BJ, Blackband SJ, Chatham JC, Mathews VP, Bryan RN. Quantitation of Proton Nmr-Spectra of the Human Brain Using Tissue Water as an Internal Concentration Reference. Nmr in Biomedicine 1993;6(1):89-94.
[6]. De Beer R, Barbiroli B, Gobbi G, Knijn A, Kugel H, Langenberger KW, et al. Absolute metabolite quantification by in vivo NMR spectroscopy: III. Multicentre 1H MRS of the human brain addressed by one and the same data-analysis protocol. Magn Reson Imaging 1998;16(9):1107-11.
[7]. Hatami N, Sdika M, Ratiney H. Magnetic resonance spectroscopy quantification using deep learning. International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer; 2018:467-75.
[8]. Gurbani SS, Schreibmann E, Maudsley AA, Cordova JS, Soher BJ, Poptani H, et al. A convolutional neural network to filter artifacts in spectroscopic MRI. Magn Reson Med 2018;80(5):1765-75.
[9]. Lee HH, Kim H. Intact metabolite spectrum mining by deep learning in proton magnetic resonance spectroscopy of the brain. Magn Reson Med 2019;82(1):33-48.
[10]. Lee HH, Kim H. Deep learning-based target metabolite isolation and big data-driven measurement uncertainty estimation in proton magnetic resonance spectroscopy of the brain. Magn Reson Med 2020;n/a(n/a).
[11]. Chandler M, Jenkins C, Shermer S, Langbein F. MRSNet: Metabolite Quantification from Edited Magnetic Resonance Spectra With Convolutional Neural Networks. arXiv preprint arXiv:190903836 2019.
[12]. Birch R, Peet AC, Dehghani H, Wilson M. Influence of macromolecule baseline on (1) H MR spectroscopic imaging reproducibility. Magn Reson Med 2017;77(1):34-43.
[13]. Seeger U, Klose U, Mader I, Grodd W, Nagele T. Parameterized evaluation of macromolecules and lipids in proton MR spectroscopy of brain diseases. Magn Reson Med 2003;49(1):19-28.
[14]. Opstad KS, Bell BA, Griffiths JR, Howe FA. Toward accurate quantification of metabolites, lipids, and macromolecules in HRMAS spectra of human brain tumor biopsies using LCModel. Magn Reson Med 2008;60(5):1237-42.
[15]. Lee HH, Kim H. Parameterization of spectral baseline directly from short echo time full spectra in (1) H-MRS. Magn Reson Med 2017;78(3):836-47.
[16]. Provencher SW. LCModel & LCMgui user’s manual. 2019.
[17]. Yamano T. Lecture 4: Multivariate Regression Model in Matrix Form 2010.
[18]. Greene WH. Econometric analysis. fifth ed.: Pearson Education India; 2003.
[19]. Wu Ty. Investigating the reproducibility on the quantification of γ-aminobutyric acid in visual cortex. Department of Electronic and Computer Engineering. National Taiwan University of Science and Technology; 2013.
[20]. Träber F, Block W, Lamerichs R, Gieseke J, Schild HH. 1H metabolite relaxation times at 3.0 tesla: Measurements of T1 and T2 values in normal brain and determination of regional differences in transverse relaxation. J Magn Reson Imaging 2004;19(5):537-45.
[21]. Choi C, Coupland N, Kalra S, Malykhin N, Bhardwaj P, Gheorghiu D, et al. Brain Glutamate Proton Transverse Relaxation Time as measured by Spectrally-Selective Refocusing. 2006.
[22]. Edden RA, Pomper MG, Barker PB. In vivo differentiation of N‐acetyl aspartyl glutamate from N‐acetyl aspartate at 3 Tesla. Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine 2007;57(6):977-82.
[23]. Dhamala E, Abdelkefi I, Nguyen M, Hennessy TJ, Nadeau H, Near J. Validation of in vivo MRS measures of metabolite concentrations in the human brain. NMR Biomed 2019;32(3):e4058.
[24]. Hädel S, Wirth C, Rapp M, Gallinat J, Schubert F. Effects of age and sex on the concentrations of glutamate and glutamine in the human brain. J Magn Reson Imaging 2013;38(6):1480-7.
[25]. Chiu PW, Mak HK, Yau KK, Chan Q, Chang RC, Chu LW. Metabolic changes in the anterior and posterior cingulate cortices of the normal aging brain: proton magnetic resonance spectroscopy study at 3 T. Age (Dordr) 2014;36(1):251-64.