Auditory Data Pattern Analysis for Differential Diagnosis of Hidden Hearing Loss

Abstract

As smartphones became essential to our daily lives, hearing loss and tinnitus patients are also growing continuously. Tinnitus is a symptom feeling phantom sound like ringing, buzzing, clicking. Some tinnitus cases do not disturb daily life, but severe tinnitus patients report to the discomfort and have side effects like insomnia, dyspepsia, etc. Much research tried to find the relationship between tinnitus and hearing loss. According to animal research results, hearing loss can be developed after tinnitus, and sensorineural hearing loss cases are detected with inner hair cell degeneration. Moreover, hidden hearing loss is reported recently, and many follow-up research was conducted to find the cause and process of hidden hearing loss. Hidden hearing loss occurs when a temporal degeneration of the inner hair cell or synapse terminal is exposed to loud noise. It takes few days to recover the original statement or cannot recover if the degeneration is severe. So, hidden hearing loss can be classified as sensorineural hearing loss in a large category. History taking research with human audiometry data for sensorineural hearing loss and tinnitus was executed to find the human case relationship. Statistical analysis is performed for finding the patterns of tinnitus patients with the normal group, comparing unilateral tinnitus patients for ipsilateral ear and contralateral ear. Decrease of amplitude in ABR wave I, prolonged latency in wave I, wave V, different latency interpeak pattern, amplitude ratio, and relationship with PTA frequencies are the results of history-taking. However, such a history-taking result is an analysis with few patient reports, which was a record when the patients did audiometry. In this study, approximately 11,200 ABR data and 9,060 PTA data are used with big data analysis. At first, ABR data was used to find the patterns for several cases, bilateral hearing loss cases, unilateral hearing loss cases, bilateral hearing loss with tinnitus cases, bilateral hearing loss with unilateral tinnitus cases, bilateral tinnitus with normal hearing cases. Secondly, found the PTA data pattern for conductive hearing loss and sensorineural hearing loss. Sensorineural hearing loss cases were divided into two types, high-frequency hearing loss and low-frequency hearing loss. Target data is regarded as a hidden hearing loss sufficient for the ABR data pattern and shows low-frequency hearing loss patterns. Non-target data shows the pattern of high-frequency hearing loss. Logistic regression (LR), support vector classifier (SVC), random forest classifier (RF) are used for binary decision validation of target data. The train dataset consists of sensorineural hearing loss cases, and test data consists of target and non-target data. The gap value of all possible cases in PTA was used for data value. The validation result shows meaningful accuracy (RF:0.85, LR:0.89, SVC:0.92) for prediction, and found some featured data with useful parameters. For further study, research should prove the meaning of the featured dataset and make an improved algorithm to diagnose the possibility of hearing loss with simple audiometry.