Adaptive noise depression for functional brain network estimation.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Psychiatry

Background:

• Autism spectrum disorder (ASD) diagnosis relies on behavioral symptoms, hindering early detection and intervention.
• Functional brain networks (FBNs) offer potential biomarkers for early neuro-disease diagnosis but are often inaccurately estimated due to noise.
• Existing FBN estimation methods inadequately address noise, compromising diagnostic accuracy.

Purpose of the Study:

• To develop a novel approach for estimating FBNs that explicitly models and accounts for noise during data acquisition.
• To enhance the accuracy and reliability of FBNs for improved early diagnosis of neurological disorders, specifically ASD.
• To create a flexible, plug-and-play noise module adaptable to various FBN estimation techniques.

Main Methods:

• Introduced a noise modeling framework incorporating a noise term to represent errors and a regularizer for noise prior.
• Formulated the method as FBN estimation on transformed fMRI data, enabling modification of traditional techniques.
• Developed a "plug-and-play" noise module for integration into existing FBN estimation methods.

Main Results:

• Demonstrated improved FBN quality through noise modeling.
• Achieved up to a 13.04% improvement in classification accuracy for identifying ASD from normal controls (NCs) compared to baseline methods.
• Validated the effectiveness and flexibility of the proposed noise modeling approach in experimental settings.

Conclusions:

• The proposed noise modeling framework significantly enhances FBN estimation for neurological disorders.
• This approach offers a flexible and effective tool for improving early ASD diagnosis through more accurate FBN analysis.
• The "plug-and-play" noise module provides a practical advancement for existing neuroimaging analysis pipelines.