Cortical surface thickness as a classifier: boosting for autism classification.

Area of Science:

• Neuroimaging
• Machine Learning
• Developmental Neuroscience

Background:

• Classifying autism spectrum disorder (ASD) from control groups using structural neuroimaging is challenging due to complex brain surface topology.
• Cortical thickness is a promising neuroimaging biomarker, but its high dimensionality and small training datasets create difficult learning problems.

Purpose of the Study:

• To develop a novel algorithm for classifying autistic individuals from controls using only structural brain image data.
• To address the challenges of feature extraction and high-dimensional data in ASD classification.
• To incorporate topological priors into the classification framework to identify discriminative brain regions.

Main Methods:

• Proposed a new LPBoosting algorithm with truncated neighborhoods, treating each point on the brain's cortical surface as a hypothesis.
• Incorporated topological priors, ensuring neighboring points on the cortical surface (hypothesis pairs) have similar discriminative qualities.
• Utilized structural image data, specifically cortical thickness, for classification.

Main Results:

• Successfully learned a high-quality classifier for distinguishing between autistic and control groups.
• The algorithm provided not only classification labels but also identified specific "discriminative regions" on the cortical surface.
• Demonstrated the effectiveness of incorporating topological priors into the classification framework.

Conclusions:

• The proposed LPBoosting algorithm offers a robust method for ASD classification using structural neuroimaging.
• Identifying discriminative regions enhances the interpretability of the classification model.
• This approach advances the use of neuroimaging biomarkers for clinical diagnosis of ASD.