利用靜息態功能磁振影像對注意力不足過動症患者進行機器與深度學習分類

Machine and Deep Learning Classification of Patients with Attention Deficit Hyperactivity Disorder Using Resting-state Functional Magnetic Resonance Imaging

汪靜(Jing Wang) ； 許敦韋(Tun-Wei Hsu)

《北市醫放雜誌》 11卷1期 (2021/12) Pp. 10-21

https://doi.org/10.6663/TPART.202112_11(1).0002

摘要

注意力不足及過動症（Attention Deficit and / Hyperactivity Disorder, ADHD），主要以衝動、注意力不集中和過動等症狀，造成兒童的學習、情緒、社會關係、認知方面和適應障礙。如果可以從大腦的結構或功能影像找到有效的分類方法，可以對患者的生活產生深遠的影響，這也會導致治療方法的改進和提升。靜息態功能磁振造影（resting-state functional Magnetic Resonance Imaging, rs-fMRI）是一種廣泛使用的神經影像學工具，並使用各種演算法分析大腦的功能連結。雖然人工智能的機器和深度學習技術運用於分析rs-fMRI數據大有進展，但是研究人員如何選擇合適的機器或深度學習模型來分析rs-fMRI數據是仍是一個很大的挑戰。一個合理的原因可能是原始rs-fMR數據非常複雜，而機器學習在分類rs-fMRI數據沒有那麼強大。而深度學習它可以直接從原始數據中學習，以自動提取潛在特徵多層非線性神經網絡分類。所以本文概述並描述各種監督式機器與深度學習模型，了解是否有助研究人員分類正常發育組與ADHD患者的rs-fMR數據。設想未來可以完成其他神經退化疾病以及精神障礙的診斷，並使用磁振造影技術和機器學習模型進行量化，以提高醫師的鑑別診斷力。

關鍵字

機器學習 ； 深度學習 ； 注意力不足及過動症 ； 靜息態功能磁振造影

並列摘要

Attention Deficit and / Hyperactivity Disorder (ADHD) is characterized by symptoms such as impulsivity, inattention, and hyperactivity, cause learning, mood, social relations, cognitive and adjustment disorders in children. If we find an effective classification method from the structure or function of the brain, it can have a profound impact on the life of the patient, which will also lead to the improvement and promotion of treatment methods. Resting-state functional magnetic resonance imaging (rs-fMRI) is a widely tool that uses various algorithms to analyze the functional connections in brain. Although there has been great progress in the application of machines and deep learning techniques to analyze rs-fMRI data, it is still a great challenge for researchers to choose the appropriate learning model to analyze rs-fMRI data. A reasonable reason may be that the rs-fMR data is very complex, while machine learning is not so powerful in classifying disease from rs-fMRI data. In deep learning, it can learn directly from the original data to automatically extract potential features for multi-layer nonlinear neural network classification. So, in this paper we outline and describe various supervised machine and deep learning techniques to see which one can help researchers classify rs-fMR data of normal development group and ADHD patients. We envision that other neurodegeneration and mental disorders can be diagnosed in the future and quantified using magnetic resonance imaging technology and machine learning models to increase diagnosis accuracy.

並列關鍵字

Machine learning ； Deep learning ； Attention Deficit and Hyperactivity Disorder ； resting state functional magnetic imaging