Voice-driven Parkinson's disease multi-symptom characteristic parameter small sample learning method

Abstract

The invention relates to a speech-driven small sample learning method for Parkinson's disease multi-symptom characteristic parameters, which realizes speech analysis of Parkinson's disease patients under a small-scale data set, inputs the first-order characteristics of the speech into a convolutional neural network, combines the obtained high-order characteristics to obtain a prototype vector, and carries out speech learning on the Parkinson's disease multi-symptom characteristic parameters. Calculating the difference between the prototype and the to-be-tested voice through a distance comparison method, predicting the probability of a to-be-tested sample under each symptom, training a model through a cross entropy loss function and an Adam optimizer in deep learning, enabling a prototype vector to be close to the real distribution of the symptom of a patient, and obtaining the real distribution of the symptom of the to-be-tested sample. And taking the finally obtained prototype vector as a key feature to realize simultaneous prediction and analysis of various symptoms of the Parkinson's disease patient.

Description

technical field [0001] The invention belongs to the medical application in the field of deep learning, and relates to a small-sample learning method and system for voice-driven Parkinson's disease multi-symptom characteristic parameters. Background technique [0002] Parkinson's disease is a common chronic neurodegenerative disease, and its symptoms include motor symptoms and non-motor symptoms. According to the different movement disorders, patients can be divided into tremor dominant type (TD) and posture / gait disorder type (PIGD). Motor symptoms include tremor, freezing of gait, dysphagia, dyskinesias, on-off periods, etc. Tremor is mainly manifested as persistent or intermittent tremor of the limbs or the whole body, which is more common in tremor-dominant type; freezing gait is mainly manifested by decreased stride length or sudden stop when the patient walks, difficulty in starting, blocked legs, and difficulty in walking, which is more common in Posture / gait disorde...

AI Technical Summary

Problems solved by technology

[0004] When using speech signals to analyze the condition of Parkinson's disease, the quality of the speech must first be evaluated to determine whether the speech contains too much noise that affects the analysis results. Patients generally use ordinary smart phones to record speech, and the recording environment is generally residential or Hospitals, such an environment cannot guarantee the voice quality of each piece of data, resulting in very little voice data that can meet the quality evaluation

Images