OSAHS Screening Based On Neural Network

Obstructive sleep apnea-hypopnea syndrome(OSAHS)is a high-grade sleep-disordered respiratory disease.Patients with upper airway obstruction during sleep can easily cause suffocation or even sudden death.In hospital,polysomnography can detect OSAHS.However equipment has the disadvantages of sparse number,cumbersome operation,and timeconsuming,which makes most patients fail to be detected in time.A convenient,fast,and suitable method for home use detection is currently urgently needed.This paper propose a OSAHS screening method based on deep learning neural network model.Through deep learning neural network to analyze the snoring data of OSAHS patients,the potential characteristic information in the snoring data is excavated.The study of this article is based on the clinical features of OSAHS disease.Because the upper airway structure of OSAHS patients is obviously different from ordinary people,an effective measure to detect the airway obstruction in patients is to detect whether the patient's snoring signal is abnormal.First,the snoring data is extracted and processed using speech signal processing technology.The processed snoring signal is then sent to the Sparse Autoencoder neural network for feature extraction to find the difference between OSAHS snoring and normal snoring.Finally using the extracted features to complete the OSAHS recognition.Because there are individual differences in the snoring signal,and the original snoring signal contains more redundant information,this screening method has deficiencies.Therefore,this paper proposes a OSAHS screening method based on the Sparse Autoencoder neural network based on the snore feature.Mel cepstrum coefficient feature and linear prediction cepstrum coefficient feature of the processed snoring acoustic signal are extracted.These two characteristics are respectively used as the input of the Sparse Autoencoder neural network,so that the snoring signal is after entering the Sparse Autoencoder neural network,a redundant information is moved;thus,the Sparse Autoencoder neural network can be better used to analyze the potential features existing in the click signal.In addition,the corresponding experimental simulation was carried out in this paper to verify the feasibility of the method.The experiment selected typical snore of OSAHS patients as a training sample set,and collected snoring data of 54 subjects as a test sample set,including 27 normal snorers and 27 OSAHS snorers.In this paper,the Mel cepstrum coefficient feature and the linear predictction cepstrum coefficient feature of the snore signal are respectively taken as the input of the Sparse Autoencoder neural network,and then the OSAHS screening is realized by the output feature.The screening results based on the Mel cepstrum obtained 81%,88% of the screening sensitivity and specificity;the screening results based on the characteristics of the linear prediction cepstrum coefficient obtained 92%,88% of the screening sensitivity and specificity.The experiment verifies the feasibility and correctness of this this method,and meets the requirements of OSAHS in clinical medicine.