使用適應性演算法之磁振造影主動抗噪系統

Abstract

磁共振影像(Magnetic Resonance Imaging)在醫學影像上有卓越的貢獻，然而在取得影像的過程中，卻會產生100分貝以上的噪音，其中以回波平面影像序列(Echo-planar imaging, EPI)為最。對於這樣的音量，人體只能承受約30分鐘的掃描時間。本研究是討論如何降低這些噪音。 目前已有許多論文致力解決磁共振影像的噪音問題，像是透過改善磁場線圈、設置真空層隔絕介質等方法減少噪音。但這些方式皆過於昂貴，且不能在現有的醫療器材上使用。較為實際的方法為使用耳塞、耳罩等被動材料消除高頻噪音，再加上主動式噪音抑制。即利用耳機主動製造出與噪音波型正負相反的波，稱之為抗噪波，藉由聲波的正負相消來抗噪，消除低頻噪音。 先前的研究大多會使用FxLMS演算法來進行降噪，但是效果不甚佳，仍需改善。本研究針對EPI產生的噪音設計了一種新的適應性演算法，利用EPI所產生重複性噪音的特性，將先驗資訊納入考量，稱為先驗資訊演算法(Prior information Based Algorithm)。此演算法能準確預測噪音的波型，並有效的地播放出抗噪波做抵消，來達成抗噪目的。而本實驗也將比較FxLMS演算法和先驗資訊演算法於模擬和實作上的優劣。 本實驗除了提出新的演算法構想，以及電腦模擬，更實構了一個主動式噪音抑制的系統。已經實際於本實驗室的3T(Tesla)磁共振影像系統上有26分貝的總降噪效果。其中被動式抗噪約16分貝，主動式抗噪約10分貝。在未來，可以將此系統實際應用於功能性磁振造影(functional MRI)系統中，讓受試者在更舒服的環境進行心理學實驗，尤其是需要降低噪音的聽覺刺激或語音刺激方面的研究。

Magnetic Resonance Imaging (MRI) is an important modality in medical care systems. However, MRI conducts the noise over 100dB while acquiring the imaging, especially for Echo-Planar Imaging (EPI). Subjects or Patients will feel uncomfortable after half of an hour under this noisy environment. Such that, reducing the noise level and providing the comfortable environment is a big issue for MRI scanning. The traditional methods to reduce the noise is limited, such as providing the passive material (ex. Ear plug) is failed to reduce the low frequency noise, or improving the MRI system (ex. change MRI scanning sequence and implement the vacuum layer in the MRI system) is hard to implement on the current system. To solve the noise problem, the headset with active noise cancellation was established. Active noise cancellation was using a headset to produce an inverse waveform in order to cancel the noise waveform conducted from the system, especially the low-frequency noise. A novel algorithm required the EPI noise as the prior information was used in generate the inverse waveform, in which the periodic properties of EPI noise was considered, so called Prior information Based Algorithm. The pre-required prior information was considered as a template in order to predict the following EPI noise. Our approach was combined the active noise cancellation and passive material, not only provided the low frequency noise reduction in a cost effective way but also compatible to the current systems. In this study, we also compare FxLMS algorithm with Prior information based algorithm in simulation and realistic results. Besides providing the novel idea of algorithm and computer simulation, a real noise cancellation system has been demonstrated in 3T MRI system. The noise cancellation system revealed 26dB noise reduction totally, whereas the passive reduction was 16dB and active reduction was 10dB. In the future, the noise cancellation system will be applied for the MRI scanning to provide a more comfortable environment, especially in the fMRI study with auditory task.