[Analysis of related factors between sudden sensorineural hearing loss and serum indices base on artificial intelligence and big data]

Abstract

Objective:The etiology and pathophysiologic mechanism of sudden sensorineural hearing loss are undefined. We will use artificial intelligence and big data methods to explore the correlation between sudden sensorineural hearing loss and serum indices. Method:A total of 1218 patients with sudden deafness admitted to Sun Yat-sen Memorial Hospital were selected as the experimental group, 95 861 healthy subjects were randomly selected as the control group at the same period. Serum biochemical indexes in two groups were collected and analyzed by TreeNet and CART machine learning algorithms, to screen out highly correlated indicators with sudden sensorineural hearing loss and dig out a set of clinical features for people with high risk of sudden sensorineural hearing loss. Result:It was found that high prevalence rate of sudden sensorineural hearing loss is related to eosinophils, reticulocyte and fibrinogen. The areas under the receiver operator characteristic curves（ROC-AUC） were exploited to evaluate the prediction performance of TreeNet model. Overall the TreeNet model has provided high predictive ability by ROC curve, achieving AUC of 0.99, both recall and accuracy rate of 99.90%. Conclusion:There is significant difference between sudden deadness and normal people in serum biochemical indexes. Eosinophil is the first important indicator to distinguish sudden sensorineural hearing loss. Treenet model has important referenced significance for the screening and diagnosis of sudden sensorineural hearing loss.

Keywords: CART; Sudden sensorineural hearing loss; TreeNet; artificial intelligence; machine learning.