Assessing Predictive Ability of Dynamic Time Warping Functional Connectivity for ASD Classification.

Area of Science:

• Neuroimaging
• Machine Learning
• Developmental Neuroscience

Background:

• Functional connectivity MRI (fcMRI) measures brain region correlations using blood oxygen-level-dependent (BOLD) signals.
• Traditional Pearson correlation (PC) assumes no time lag between BOLD signals, potentially missing complex temporal dynamics.
• Autism spectrum disorder (ASD) diagnosis can benefit from advanced neuroimaging analysis techniques.

Purpose of the Study:

• To evaluate Dynamic Time Warping (DTW) fcMRI as an alternative to PC fcMRI for classifying ASD.
• To compare the predictive performance of DTW fcMRI and PC fcMRI using machine learning models.
• To investigate optimal cross-validation strategies for machine learning models in neuroimaging studies.

Main Methods:

• Collected fcMRI data using both PC and DTW measures.
• Employed machine learning models with dimension reduction techniques (e.g., principal component analysis) for ASD classification.
• Assessed various cross-validation (CV) methods, including K-fold nested within leave-one-out CV.

Main Results:

• DTW fcMRI demonstrated superior predictive ability compared to PC fcMRI when combined with dimension reduction.
• DTW fcMRI captures distinct, potentially complementary, functional connectivity patterns compared to PC fcMRI.
• Nested K-fold within leave-one-out CV offers a competitive balance of performance and computational efficiency for small sample sizes.

Conclusions:

• DTW fcMRI is a promising alternative for analyzing brain connectivity and holds potential for improved ASD classification.
• DTW fcMRI provides complementary information to PC fcMRI, warranting further investigation.
• Optimized cross-validation strategies are crucial for robust machine learning model development in neuroimaging research.