Automatic sleep staging algorithm for stochastic depth residual networks based on transfer learning / 生物医学工程学杂志

The existing automatic sleep staging algorithms have the problems of too many model parameters and long training time, which in turn results in poor sleep staging efficiency. Using a single channel electroencephalogram (EEG) signal, this paper proposed an automatic sleep staging algorithm for stochastic depth residual networks based on transfer learning (TL-SDResNet). Firstly, a total of 30 single-channel (Fpz-Cz) EEG signals from 16 individuals were selected, and after preserving the effective sleep segments, the raw EEG signals were pre-processed using Butterworth filter and continuous wavelet transform to obtain two-dimensional images containing its time-frequency joint features as the input data for the staging model. Then, a ResNet50 pre-trained model trained on a publicly available dataset, the sleep database extension stored in European data format (Sleep-EDFx) was constructed, using a stochastic depth strategy and modifying the output layer to optimize the model structure. Finally, transfer learning was applied to the human sleep process throughout the night. The algorithm in this paper achieved a model staging accuracy of 87.95% after conducting several experiments. Experiments show that TL-SDResNet50 can accomplish fast training of a small amount of EEG data, and the overall effect is better than other staging algorithms and classical algorithms in recent years, which has certain practical value.

Decomposição e análise dos potenciais evocados auditivos de tronco encefálico / Decomposition and analysis of auditory brainstem response

Os PEATEs são sinais resultantes da combinação de respostas de atividades neurais a estímulos sonoros no córtex. Caracteriza-se por ondas, sendo seus picos nomeados por algarismos romanos (I, II, III, IV, V, VI e VII). O processo clássico de identificação desses picos é baseado na visualização do sinal gerado pela promediação de cada amostra. Nele são identificadas as características morfológicas do sinal e os aspectos temporais relevantes constituídos pelas ondas de Jewett no qual cada onda tem uma relação anatômica com o sítio de origem. No entanto, durante esse processo de identificação visual surgem dificuldades que tornam a análise visual dos PEATE uma fonte constante de dúvidas em relação a fidedignidade e concordância de marcação dos picos pela subjetividade entre os examinadores. Com o objetivo de melhorar o processo de avaliação dos PEATE, foi desenvolvido um sistema de detecção automática para os picos, com capacidade de aprendizado que leva em consideração o perfil de marcação prévia realizado por examinadores, podendo ser considerado também, as marcações futuras de examinadores que utilizarão o software como auxílio em suas análises. Para a detecção de picos foi utilizada a Transformada Wavelet Contínua, associada a um Classificador Probabilístico construído a partir de marcações realizadas pelos examinadores. Para a avaliação do sistema foram utilizadas 748 amostras de PEATE de 11 sujeitos. A avaliação do sistema proposto apresentou uma taxa de acerto 74,3% a 99,7%, entre o sistema e a marcação manual, de acordo com o tipo de onda analisada. O presente estudo foi concebido com a intenção de ser uma ferramenta prática e por isso voltada para a aplicação clínica. Os resultados apresentados mostram uma técnica eficaz e capaz de aperfeiçoar o processo de avaliação dos PEATEs. A técnica proposta se mostra precisa mesmo na presença de ruído, característico de sinais biológicos especialmente no PEATE por ser um sinal de amplitude baixa.

Auditory Brainstem Response (ABR) results from the combination of neural activity responses in the presence of sound stimuli, detected by the cortex and characterized by peaks and valleys. They are identified by Roman numerals (I, II, III, IV, V, VI and VII). The identification of these peaks is carried out by the classical manual process of analysis, which is based on the visual/manual processing of the signals. The morphological and temporal characteristics of the signal carry relevant physiological and anatomical information regarding the auditory system. However, in this visual process of analysis some difficulties may occur, specifically, the results of the analysis may vary according to the type of protocol, settings of equipment employed, and the experience of the examiner. This makes the analysis of ABR subject to the influence of many variables that may interfere on the reliability and agreement of results obtained in distinct research centers and by different examiners. Therefore, the main propose of this study was to develop and assess a system capable of automatically detecting and classifying ABR waves, which are called Jewett waves. A relevant feature of the proposed tool is that it can learn from the experience of examiners continuously. In order to evaluate the system approximately 748 samples of ABR obtained from 11 subjects were analyzed by the automatic system. These results were compared to analyses obtained from five seasoned examiners, and they showed a high level of agreement, ranging for 74.3% to 99.7%, between responses given by the system and the examiners. Thus the proposed technique is proved to be accurate even in the presence of noise, especially characteristic of the ABR that is a sign of low amplitude.