Supervised Approach to Identify Autism Spectrum Neurological Disorder via Label Distribution Learning

Area of Science:

• Neuroscience
• Computational Biology
• Medical Informatics

Background:

• Autism Spectrum Disorder (ASD) presents complex neurodevelopmental challenges.
• Existing classification algorithms struggle with ASD data characteristics like label noise, high dimensionality, and class imbalance.
• Current diagnostic systems are often binary and fail to capture the spectrum's complexity.

Purpose of the Study:

• To develop an improved classification method for Autism Spectrum Disorder (ASD).
• To address limitations in existing algorithms, specifically data label noise and sample imbalance.
• To enhance the accuracy and effectiveness of ASD diagnosis.

Main Methods:

• Utilized Label Distribution Learning (LDL) to manage noisy data labels.
• Employed Support Vector Regression (SVR) for handling imbalanced sample data.
• Implemented a cost-sensitive approach to correct sample imbalance.
• Applied LDL to overcome high-dimensional feature classification challenges by mapping samples to a feature space for multiclass ASD diagnosis.

Main Results:

• The proposed method effectively balances the influence of majority and minority classes.
• Demonstrated significant improvement in handling imbalanced data crucial for ASD diagnosis.
• Outperformed previous methods in classification performance and accuracy.
• Successfully resolved issues related to unbalanced data in ASD diagnostic contexts.

Conclusions:

• The novel approach offers a more robust solution for ASD diagnosis by tackling inherent data complexities.
• This method enhances diagnostic accuracy and reliability for Autism Spectrum Disorder.
• The strategy provides a valuable tool for improving the classification and understanding of neurodevelopmental disorders like ASD.