Research On Gradient Noise Cancellation In Mri Scanning

As an effective non-invasive approach to acquire the diagnostic image, Magnetic resonance imaging (MRI) has become one of the most important modern medical diagnostic tools. It plays an irreplaceable role in clinical diagnosis and research. The MRI technique is based on the resonance phenomenon of nucleus in a magnetic field. In order to obtain the location information and depict a spatial distribution image of the proton, it needs three additional gradient magnetic field generated by the gradient coils. However, when the current in the gradient coil rapidly change, the gradient coil itself will vibrate under the action of Lorentz force in the magnetic field, making loud noise.The gradient noise is harmful to the human health. In some functional MRI it often makes a bad effect to the quality of the image. It is a serious impediment of listening to the doctor and the patient in the scan room as well. MRI gradient noise elimination has been one of the main directions of efforts of the MRI device manufacturers and scientists from various countries. This study has a practical theoretic value and broad market prospects.In this paper, the principle of MRI and the sequence parameters are introduced at the first. The characteristic of the gradient noise, particularly, the relationship with the sequence parameters is examined to check if there is any help to noise cancellation. The main efforts are pay to the research on the noise reduction method for the gradient noise and the job of porting it to TI's DSP hardware for real-time processing.Sound pressure level of the gradient noise is extremely high, and its energy is mainly between 1 KHz and 3 KHz, that is overlapped with the human voice spectrum regions. Patient's voice will be submerged in the gradient noise. However, the gradient noise is highly repeatable, and has a direct relationship with the sequence. Highly discriminable spectrum lines can be observed in the frequency domain of the gradient noise scanning TSE and GRE sequence, and its interval depends on the sequence parameters such as TR. A spectral subtraction method with a direct reference noise is proposed and tested for the gradient noise, which is based on the adaptive noise cancellation method and the normal spectral subtraction speech enhancement method. It has an obvious effect though the gradient noise is in low SNR. The algorithm is then implemented on the TMS320C6416 digital signal processor. A signal processing framework is setup based on a RTOS DSP/BIOS. The real-time requirement is fully met.Practical test shows that there is more than 20 dB SNR improvement calculated with the minimum statistics method, resulting in a subjective feeling of significant improvement.