ABSTRACT
This thesis, with a Class 3A hospital as the research object, aims at analyzing the status quo and causes of its costs of health materials so as to put forward corresponding methods and suggestions of cost management, thus providing scientific guidance on reducing operating costs and improve medical quality in public hospitals.
ABSTRACT
A new approach is put forward for reducing the number of trials required for the extraction of the brain event related potentials (ERPs). The approach is developed by combining both the subspace methods and lift wavelet transform. First, the signal subspace is estimated by applying the singular value decomposition (SVD) to an enhanced version of the raw data obtained by orthonormal projection of the raw data onto the estimated signal subspace. At the same time, the colored noise is whitened. Next, the ERPs are extracted by lift wavelet construction of the enhanced version. Simulation results show that combination of both the subspace methods provides much better capability than does each of them. The experiments showed that the practical results were good.
Subject(s)
Humans , Algorithms , Brain , Physiology , Data Interpretation, Statistical , Electroencephalography , Methods , Event-Related Potentials, P300 , Physiology , Evoked Potentials , Physiology , Signal Processing, Computer-AssistedABSTRACT
We constructed a Brain-computer interface-based mental speller which realizes user-computer interaction. The feature signals of user's intention are embedded in spontaneous EEG background. Single-trial feature estimation should be used on this online occasion instead of the grand average usually used in cognitive or clinical experiments. To demonstrate this technique beyond laboratories, fewer EEG recording channels are preferred. A unique paradigm, which is called imitating-natural-reading, was exploited to induce visual evoked potentials. We explored the single-trial estimation of VEP recorded in single channel using support vector machine on three subjects, and obtained satisfactory data, the classification accuracy being 92.1%, 94.1% and 91.5%, respectively. These results put forward a significant step fowards the ultimate realization of our mental speller.