Machine Learning Exploration of Brain Morphological Features and Sensory Measures

Area of Science:

• Neuroscience
• Neuroimaging
• Machine Learning

Background:

• Identifying neuroanatomical markers for sensory systems has been challenging.
• Previous research has explored the brain basis of sensory perception, but definitive biomarkers are lacking.

Purpose of the Study:

• To investigate the relationship between brain morphology and sensory performance in audition, olfaction, taste, and visual contrast sensitivity.
• To identify significant neuroanatomical features associated with sensory abilities using machine learning.
• To develop machine learning models for classifying individuals based on sensory performance using brain imaging data.

Main Methods:

• Utilized a large dataset (n=874) from the Human Connectome Project.
• Employed machine learning techniques, including Random Forest, SelectKBest, Recursive Feature Elimination, and SHAP for feature selection.
• Applied 5-fold and leave-one-out cross-validation for model performance evaluation.
• Conducted binary classification to distinguish high vs. low sensory performance groups.

Main Results:

• Machine learning models achieved satisfactory accuracy in classifying sensory performance.
• Identified significant neuroanatomical features correlating with sensory abilities.
• Achieved 67% accuracy for olfaction and 71% for visual contrast sensitivity classification.

Conclusions:

• Machine learning integration with neuroimaging data provides novel insights into the neural underpinnings of sensory perception.
• This study highlights the potential of neuroimaging biomarkers for understanding sensory system variations.
• Further research can build upon these findings to explore the neuroanatomical basis of human senses more deeply.