[Research on Parkinson's disease recognition algorithm based on sample enhancement]

Area of Science:

• Neurology
• Artificial Intelligence
• Signal Processing

Context:

• Parkinson's disease (PD) affects vocal cord function, altering voiceprint characteristics.
• Early detection of PD is crucial for effective management.
• Existing voiceprint datasets for PD are often limited in size.

Purpose:

• To develop an advanced deep learning model for enhancing limited Parkinson's disease voiceprint datasets.
• To improve the accuracy of early Parkinson's disease recognition using enhanced speech spectrograms.

Summary:

• A double self-attention deep convolutional generative adversarial network (GAN) was proposed for generating high-resolution speech spectrograms.
• The model incorporates increased network depth, gradient penalty, and spectral normalization for enhanced sample clarity.
• A ConvNeXt classification network utilizing transfer learning was employed for feature extraction and classification.

Impact:

• The enhanced voiceprint samples demonstrated improved clarity and a better Fréchet inception distance (FID) score.
• The Parkinson's disease recognition algorithm achieved a high accuracy of 98.8%.
• This approach offers a viable solution for early Parkinson's disease diagnosis using speech analysis, particularly with small datasets.