Research Of Similarity Measurement For Biomedical Signal

Bio-signal is usually acquired through physical approaches. It describes human health status quantitatively. It can be electronic signal, or pressure signal, such as, electroencephalogram, electrocardiogram, and pulse wave etc. Diagnoses are usually depended on these changing signals. Along with de development of computer techniques, these signals were discretized, and transformed to vectors. These vectors were analyzed by various intelligent algorithm, the results were used to assist diagnosis. Since the description of similarity between signals is the key of bio-signal analysis, the accuracy of similarity measurement will influence the diagnosis directly. But, bio-signal is often accompanied by the interference of noise, baseline drift, amplitude contraction, elastic shifting of time axis, and weak in energy, traditional similarity measurement will not feasible. Through the deeply understanding of the bio-signal's characteristics, that is time series and curving, the thesis presented a similarity measurement which is suitable for bio-signal. The studies were developed in following three aspects.Under the clinical and time series features of bio-signal, the thesis found out the key point of bio-signal scientifically. Moreover, we partition the bio-signal into several segments. At last, the segments were weighted by their contribution to diagnosis. This was prepared for using in later similarity measurement function. Through the study of curving feature of bio-signal, the Hausdorff Distance was used in measuring the similarity of bio-signal. It effectively overcome those interferes which were caused by noise, baseline drift, amplitude contraction, elastic shifting of time axis, etc. And at the same time, Hausdorff Distance was modified for improving the discrimination. This is means that the key point partitioning and weight assigning on bio-signal were used in Hausdorff Distance computation. From that, the curving feature of bio-signal was taken in account. After experiment on MIT/BIH ECG database, the modified Hausdorff Distance got better discrimination.For making further efforts on the curving feature of bio-signal in similarity measurement computation, the thesis proposed a tunnel morph based similarity measurement strategy under the idea of Hausdorff Distance. Through the overlapping of class-known datasets, a graph of tunnel morph was shown. By the help of image edge detection methods, the outline of tunnel was curtained. Depended on these, a new bio-signal was classified. Experiments show that the strategy can distinguish bio-signal in various morph, and has the ability of inclusiveness. The results that proposed by the studies had broadly application value. Not only on the similarity measuring of bio-signal, but on the establishing of bio-signal class model base, clustering result evaluation of bio-signal, and human identification based on bio-signal. Except this, the studies also provided the idea and experiment methods of morph classification on the data with time series and non-linear curving feature, and proposed new idea on the further study of bio-signal.