Automatic Recognition of Auditory Brainstem Response Characteristic Waveform Based on Bidirectional Long Short-Term Memory

Frontiers in Medicine

article

Cheng Chen¹  Fei Ji²  Qiuju Wang²  Ning Yu²  Ruoxiu Xiao¹  Li Zhan²  Xiaoxin Pan¹  Zhiliang Wang¹  Xiaoyu Guo¹  Handai Qin²  Fen Xiong²  Wei Shi²  Min Shi² 

[1] School of Computer and Communication Engineering, University of Science & Technology Beijing;College of Otolaryngology Head and Neck Surgery, National Clinical Research Center for Otolaryngologic Diseases, Key Lab of Hearing Science, Ministry of Education, Beijing Key Lab of Hearing Impairment for Prevention and Treatment, Chinese PLA General Hospital;Institute of Artificial Intelligence, University of Science and Technology Beijing

关键词: auditory brainstem response;    characteristic waveform recognition;    neural network model;    bi-directional long short-term memory;    wavelet transform;   

DOI  :  10.3389/fmed.2020.613708

学科分类：社会科学、人文和艺术（综合）

来源: Frontiers

【 摘 要 】

Background: Auditory brainstem response (ABR) testing is an invasive electrophysiological auditory function test. Its waveforms and threshold can reflect auditory functional changes in the auditory centers in the brainstem and are widely used in the clinic to diagnose dysfunction in hearing. However, identifying its waveforms and threshold is mainly dependent on manual recognition by experimental persons, which could be primarily influenced by individual experiences. This is also a heavy job in clinical practice. Methods: In this work, human ABR was recorded. First, binarization is created to mark 1,024 sampling points accordingly. The selected characteristic area of ABR data is 0–8 ms. The marking area is enlarged to expand feature information and reduce marking error. Second, a bidirectional long short-term memory (BiLSTM) network structure is established to improve relevance of sampling points, and an ABR sampling point classifier is obtained by training. Finally, mark points are obtained through thresholding. Results: The specific structure, related parameters, recognition effect, and noise resistance of the network were explored in 614 sets of ABR clinical data. The results show that the average detection time for each data was 0.05 s, and recognition accuracy reached 92.91%. Discussion: The study proposed an automatic recognition of ABR waveforms by using the BiLSTM-based machine learning technique. The results demonstrated that the proposed methods could reduce recording time and help doctors in making diagnosis, suggesting that the proposed method has the potential to be used in the clinic in the future.

【 授权许可】

CC BY