Study On Data Preprocessing Of Electrode Interference In Brain Electrical Impedance Tomography

Cerebrovascular disease is a huge threat to human health with acute pathogenesis and high lethality rate. Severe head injury is also one of the clinical emergencies, which is a serious harm on patients’ life with secondary bleeding. Clinical studies have shown that advanced diagnosis and treatment is the key to a good prognosis and survival. So there is a need for a dynamic mo nitoring technique to implement advanced diagnosis for potentialvictims of cerebrovascular disease and severe head injury. While traditional medical imaging examinations(CT and MRI) can focus the nidus exactly, it is not a proper real-time medical monitor means for its enormousness, costly expense and radiation exposure. Although physiologic parameters monitor is fit for continuous monitoring, it’s too insensitive to monitor the early onset of the disease. In this way, there is a need for a continuously monitoring technique to guide the clinical therapy.Electrical impedance tomography(EIT) is a new type of biomedical functional imaging technique. It has several advantages like noninvasive, radiation free, low cost, functional imaging and et al. This technique has a bright future in clinical application for continuous monitoring. Research on brain EIT has been started for almost twenty years at home and abroad. The Fourth Military Medical University EIT group has resolved a series of technique difficulties in dynamic EIT system and takes the lead in the research in brain EIT clinical application. C linical experiments have proved the feasibility of brain EIT monitoring. However, there is some electrode problems occurred in previous studies. In clinical monitoring, completely or partially disconnected electrodes are common occurrence because patients’ movement and manipulation of clinical staff. Thus the data acquired will be invalid for image reconstruction. Such condition should be probably handled in order to improve brain EIT clinical application. This thesis carries out researches aiming at electrode interference preprocessing as follows:(1)Automatic real-time detection of faulty electrodesDuring long-term brain EIT monitoring, there are many factors probably affect electrode contact status, especially patients’ movement and manipulation of medical staff. This will lead to frustrated imaging results. Analyzing previous clinical data, we found that measuring channels including faulty electrodes are significantly different from the normal ones. And data acquired is little correlated with normal ones in curve shape while such electrode acts as negative drives. In this way this paper presents a real time detection algorithm for multiple error electrodes based on weighted correlation coefficient and wavelet decomposition. Both physical model data and human data show that this method is capable of accurate multiple detection.(2)Compensation for invalid data after occurrence of faulty electrodesBrian EIT monitoring is so long a procedure that it’s impossible to have one full-time staff keep watching the screen. If faulty electrodes happen, the medical staff won’t know last normal monitoring image even if such breakdown is removed quickly. In th is condition previous monitor procedure will be in vain and the validity of dynamic EIT will be brought down greatly. Because of this background, this paper proposes a series model combined with grey model and BP neural networks to compensate the invalid data. This method uses grey modeland a three- layer BP neural networks with one input and one output layer compensate the lost data. Grey Model(1,1) is good at trend forecasting, but poor in anti- noise. BP neural networks are capable of nonlinear approximat ion, which can remove random components in the dataset. Therefore combined method of these two techniques is able to interpolate missing data. Physical model and human experiments both testify that the combination is a feasible to reserve imaging informatio n effectively before electrode trouble happens.(3)Movement artifacts reduction using cubic natural spline fitting differenceIn previous clinical experiments, there is another electrode fault caused by patients’ movement and medical staff’s manipulation occurs. The impacted electrodes keep a relatively good contact so they won’t be identified as faulty ones. But the data collection has a huge mutation that generates serious movement artifacts on the reconstructed image. In this article we take advantage of cubic natural spline curve fitting to subtract abnormal signal from original dataset. During the experiment we consider movement artifacts, which show as baseline mutation, as step noise and impulse noise, and try to fit the noise using cubic natural spline curve. Then we subtract the curve fitting parts from original dataset and reconnect data series by parallel shift. Both physical model and human experiments’ results show the viability of fitting differential with remarkable reduction of signal mutation and restore of reconstructed image.In conclusion, this thesis carries out a series of researches on electrodes interference preprocessing:(1) presents a detection method for multiple faulty electrodes based on weighted correlation coefficient and wavelet decomposition;(2) propose a compensation method for erroneous electrode data using grey model and BP neural network series prediction system;(3) bring out a differential denoised method based on cubic spline fitting to reduce movement artifacts.Physical model experiments and human experiments prove they are effective preprocessing technique which can improve brain EIT clinical application in the future.