Data imputation and compression for Parkinson's disease clinical questionnaires.

Area of Science:

• Biomedical Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Medical questionnaires are crucial for data collection but present analytical challenges due to size and missing values.
• Incomplete or large datasets complicate statistical and machine learning analyses in biomedical research.

Purpose of the Study:

• To develop a novel method for simultaneous non-linear data imputation and dimensionality reduction of medical questionnaire data.
• To address the challenges posed by missing values and data size in biomedical data science.

Main Methods:

• A deeply-learnt residual autoencoder was proposed for integrated imputation and dimensionality reduction.
• The autoencoder's performance was analyzed concerning compression rates and missing data proportions.
• The method was evaluated using clinical questionnaires from the Parkinson's Progression Markers Initiative (PPMI) database.

Main Results:

• The proposed autoencoder effectively performs non-linear data imputation and dimensionality reduction.
• Performance dynamics were analyzed across varying compression rates and missing data percentages.
• The deep learning approach demonstrated superior performance compared to linear methods.

Conclusions:

• The residual autoencoder offers a robust solution for analyzing complex medical questionnaire data.
• This method enhances data representation for statistical and machine learning applications in biomedical research.
• The approach shows significant potential for improving Parkinson's disease data analysis.