The Platform Construction Of Automated Spindle Detection Algorithm And Its Application In Intelligence Research

Sleep is not just a resting state but also a way of the brain and the body's self-repair, which plays a regulatory role in the memory reconstruction and hormones. Therefore, the study of sleep and its mechanism has always been an important research in academic.The role of sleep is related to the rhythms. Sleep spindle is an important rhythm, which becomes a hot topic in recent academic research. The spindle is also important in clinic, the changes of spindle density has been found especially in some diseases(such as schizophrenia, autism, epilepsy, mental retardation, sleep disorders and neurodegenerative diseases, etc.),which is because sleep spindle is produced by the interaction of some brain regions(including the thalamic reticular nucleus, thalamic neurons, hippocampus and cortex), and can effects the loops that are related to learning, behavior awakening and sensory, therefore, spindle can be used as a diagnostic biomarkers.Therefore, the detection of sleep spindle is of great importance. The traditional detect method is the human eye detection, which always been the "gold standard", however,experimental record brain electrical signal is usually large, the human eye detection has become a very dull and heavy work in the research of sleep and very time consuming,therefore people studied all kinds of time-frequency analysis methods to detect spindles.This paper implements four kinds of spindles automatic detection algorithms which have been mentioned in some published articles: RMS, two order derivative, matching pursuit(MP) and AR model. We compare the results with the "gold standard", analysis the performance and compare the advantages and disadvantages. Experimental results show that the performance of RMS algorithm is better, whose elapse time is shortest, average recall rate is higher than 72% and the average precision is higher than 64%, both reach100% in several subjects. The average recall of two order derivative reaches 80% which is the highest of the four algorithm, but the FT is too high which cause a low precision. The MP is most time consuming, AR model performance is average. In the stability of the algorithm, the RMS algorithm has the smallest variance and the most stable, and the MP algorithm has the biggest variance value and the most unstable. Finally chose the RMS algorithm to detect the spindles and take the relative analysis with intelligence.Experiment included two parts completed by forty healthy participants. The first is EEG( Electro Encephalo Gram) recording and the second is intelligence gathering. The results show that the intellectual ranges between 97 and 128, the number of spindle waves has great difference between individuals, and the difference in density is small. The number of spindles in Fz electrode is correlated with total IQ, which is also correlated with operating IQ in F4 electrode. The density of spindle in Fp1, Fp2, F3, F4, Fz electrodes is correlated with operating IQ. However, the amplitude of spindle waves are not related to all the three groups IQ values. All the results are consistent with the results of previous studies. The study in the paper lays the foundation for the integrational research on EEG,neuroimaging and behavior and provides a theoretical basis.