基于深度学习和MFCC特征的脑卒中预测方法
|
摘要
脑卒中(cerebral stroke)又称中风、脑血管意外(cerebralvascular accident,CVA),是一种急性脑血管疾病,是我国成年人群致死、致残的首位病因,具有发病率高、致残率高、死亡率高和复发率高的特点。目前,针对脑卒中的治疗手段有限,而且疗效不太理想,预防是现阶段最好的治疗措施。为了有效预防脑卒中,笔者提出了一种基于深度学习和梅尔频率倒谱系数(Mel Frequency Cestrum Coefficient,MFCC)特征的脑卒中预测。首先,通过录音设备录取脑卒中患者和正常人的一小段特定语音;其次,对特定的语音做信号预处理,经预处理后对语音进行相应的梅尔变换,通过离散余弦变换获得MFCC语音特征;最后,将MFCC特征放入卷积神经网络进行模型训练,获取脑卒中的预测评价。实验结果表明,通过将MFCC特征输入到卷积神经网络进行模型训练,在预测准确性和鲁棒性方面具有较好表现。
Cerebral stroke, also known as stroke and cerebralvascular accident(CVA), is an acute cerebrovascular disease. It is the first cause of death and disability in adults in China. It has the characteristics of high morbidity, high disability, high mortality and high recurrence rate. At present, the treatment of stroke is limited, and the effect is not ideal. Prevention is the best treatment at this stage. In order to effectively prevent stroke, a stroke prediction method based on in-depth learning and Mel Frequency Cestrum Coefficient(MFCC) features is proposed. Firstly, a small segment of specific voice of stroke patients and normal people is recorded by recording equipment; secondly, the specific voice is pretreated by signal, and then the corresponding Mel transform is performed on the voice after pretreatment, and the MFCC voice features are obtained by discrete cosine transform; finally, the MFCC features are put into convolutional neural network for model training to obtain the prediction and evaluation of stroke. The experimental results show that MFCC features are input into the convolution neural network for model training, which has a good performance in prediction accuracy and robustness.
Cerebral stroke, also known as stroke and cerebralvascular accident(CVA), is an acute cerebrovascular disease. It is the first cause of death and disability in adults in China. It has the characteristics of high morbidity, high disability, high mortality and high recurrence rate. At present, the treatment of stroke is limited, and the effect is not ideal. Prevention is the best treatment at this stage. In order to effectively prevent stroke, a stroke prediction method based on in-depth learning and Mel Frequency Cestrum Coefficient(MFCC) features is proposed. Firstly, a small segment of specific voice of stroke patients and normal people is recorded by recording equipment; secondly, the specific voice is pretreated by signal, and then the corresponding Mel transform is performed on the voice after pretreatment, and the MFCC voice features are obtained by discrete cosine transform; finally, the MFCC features are put into convolutional neural network for model training to obtain the prediction and evaluation of stroke. The experimental results show that MFCC features are input into the convolution neural network for model training, which has a good performance in prediction accuracy and robustness.
引文
[1]杨科敏.认识并防治脑卒中[J].健康向导,2018(1):89.
[2]姜允申.脑卒中发生前的征兆[J].保健与生活,2015(1):31.
[3]Hermansky H.Perceptual Linear Predictive(PLP)Analysis of Speech[J].Journal of the Acoustical Society of America,1990,87(4):1738-1752.
[4]Jain A K,Prabhakar S,Hong L,et al.Filterbank-based Fingerprint Matching.[J].IEEE Trans Image Process,2000,9(5):846-59.
[2]姜允申.脑卒中发生前的征兆[J].保健与生活,2015(1):31.
[3]Hermansky H.Perceptual Linear Predictive(PLP)Analysis of Speech[J].Journal of the Acoustical Society of America,1990,87(4):1738-1752.
[4]Jain A K,Prabhakar S,Hong L,et al.Filterbank-based Fingerprint Matching.[J].IEEE Trans Image Process,2000,9(5):846-59.