Research And Implementation Of Prediction Of Cognitive Decline Progression Based On Spontaneous Speech

Alzheimer’s disease is a neurodegenerative disease that causes cognitive decline and ultimately the loss of basic living abilities.In the current study,how enable doctors to provide patients with better medical solutions to delay cognitive decline is an urgent problem to be solved.Current research is often based on image data such as MRI and PET,which is expensive and time-consuming.Therefore,it is of great significance to use patients’ spontaneous speech to predict cognitive decline and provide patients with better treatment options.This paper takes the spontaneous speech of AD patients as the research object,studies acoustic features and linguistic features and the fusion of the two features in terms of features,and then builds a machine learning model and a deep learning model on the extracted features.After training,Both can achieve effective prediction of cognitive decline progression in patients.The main research contents and work of this paper are as follows:1)In this paper,according to the characteristics of the small medical voice data set,the audio data enhancement work is carried out.This paper explores the effect of different types of acoustic features on the results,namely traditional acoustic features and acoustic spectrum.A machine learning model is built on the traditional acoustic features,and a deep learning model is built on the acoustic spectrum.When building a machine learning model,the effects of different models are compared,and experiments show that SVM has a better effect on small data sets than other models.When building a convolutional neural network,the effects of different attention mechanisms on the network performance were compared.This paper finally achieves a MeanF1 of 0.63 on both traditional acoustic features and acoustic spectrum,exceeding the effect of the baseline experiments.2)This paper uses a pre-trained model to extract linguistic features,which are sentence vectors extracted by BERT and features extracted by Wav2Vec2.0.Using a model that performs well in acoustic experiments,the experiment finally achieved a Mean-F1 of 0.67,which is equal to the baseline experiment.In this paper,the fusion experiment of acoustic features and linguistic features was carried out,and the final effect exceeded the experiments of acoustic features and linguistic features,and the Mean-F1 of 0.69 was obtained.3)Combining the above models and methods,this paper designs and implements a cognitive decline prediction system based on the spontaneous speech of AD patients.The system is based on We Chat applet and has the characteristics of convenient use.Users only need to search for the applet to answer the questions by voice,upload the audio data to the server,and then use the model trained in this paper to predict cognitive decline,and the system will finally display the prediction results to the user.