WAVELET-DOMAIN REGRESSION AND PREDICTIVE INFERENCE IN PSYCHIATRIC NEUROIMAGING

Area of Science:

• Neuroimaging
• Statistical modeling
• Psychiatry

Background:

• Brain imaging data is increasingly used in psychiatry for predictive modeling.
• Developing robust statistical methodologies is crucial for analyzing complex neuroimaging datasets.

Purpose of the Study:

• To propose and compare wavelet-domain statistical procedures for generalized linear models with image predictors.
• To develop methods for assessing the contribution of image predictors beyond scalar predictors, including permutation tests and confounding extensions.
• To evaluate the predictive power of spontaneous brain activity maps from fMRI for attention deficit/hyperactivity disorder (ADHD).

Main Methods:

• Wavelet-domain sparse principal component regression.
• Wavelet-domain sparse partial least squares.
• Wavelet-domain elastic net.
• Permutation tests for assessing predictor significance.
• Confounding assessment for functional or image predictors.

Main Results:

• The proposed wavelet-domain methods were applied to fMRI data to predict ADHD.
• The study investigated the role of confounding in ADHD prediction models.
• Results provide insights into the outcomes of the ADHD-200 Global Competition.

Conclusions:

• Statistical methodologies incorporating wavelet-domain analysis and careful confounding assessment can enhance the predictive power of neuroimaging data in psychiatry.
• These methods offer a framework for analyzing complex brain imaging data to predict psychiatric conditions like ADHD.
• Understanding confounding is critical for accurate image-based diagnostic algorithms.