[Progress in biomedical data analysis based on deep learning].

Traditional biomedical data analysis technology faces enormous challenges in the context of the big data era. The application of deep learning technology in the field of biomedical analysis has ushered in tremendous development opportunities. In this paper, we reviewed the latest research progress of deep learning in the field of biomedical data analysis. Firstly, we introduced the deep learning method and its common framework. Then, focusing on the proposal of biomedical problems, data preprocessing method, model building method and training algorithm, we summarized the specific application of deep learning in biomedical data analysis in the past five years according to the chronological order, and emphasized the application of deep learning in medical assistant diagnosis. Finally, we gave the possible development direction of deep learning in the field of biomedical data analysis in the future.

A Pulse Signal Preprocessing Method Based on the Chauvenet Criterion.

Pulse signals are widely used to evaluate the status of the human cardiovascular, respiratory, and circulatory systems. In the process of being collected, the signals are usually interfered by some factors, such as the spike noise and the poor-sensor-contact noise, which have severely affected the accuracy of the subsequent detection models. In recent years, some methods have been applied to processing the above noisy signals, such as dynamic time warping, empirical mode decomposition, autocorrelation, and cross-correlation. Effective as they are, those methods are complex and difficult to implement. It is also found that the noisy signals are tightly related to gross errors. The Chauvenet criterion, one of the gross error discrimination criterions, is highly efficient and widely applicable for being without the complex calculations like decomposition and reconstruction. Therefore, in this study, based on the Chauvenet criterion, a new pulse signal preprocessing method is proposed, in which adaptive thresholds are designed, respectively, to discriminate the abnormal signals caused by spike noise and poor-sensor-contact noise. 81 hours of pulse signals (with a sleep apnea annotated every 30 seconds and 9,720 segments in total) from the MIT-BIH Polysomnographic Database are used in the study, including 35 minutes of poor-sensor-contact noises and 25 minutes of spike noises. The proposed method was used to preprocess the pulse signals, in which 9,684 segments out of a total of 9,720 were correctly discriminated, and the accuracy of the method reached 99.63%. To quantitatively evaluate the noise removal effect, a simulation experiment is conducted to compare the Jaccard Similarity Coefficient (JSC) calculated before and after the noise removal, respectively, and the results show that the preprocessed signal obtains higher JSC, closer to the reference signal, which indicates that the proposed method can effectively improve the signal quality. In order to evaluate the method, three back-propagation (BP) sleep apnea detection models with the same network structure and parameters were established, respectively. Through comparing the recognition rate and the prediction rate of the models, higher rates were obtained by using the proposed method. To prove the efficiency, the comparison experiment between the proposed Chauvenet-based method and a Romanovsky-based method was conducted, and the execution time of the proposed method is much shorter than that of the Romanovsky method. The results suggest that the superiority in execution time of the Chauvenet-based method becomes more significant as the date size increases.