Effects of different preprocessing factors on radiomics-based morphological brain networks.

Area of Science:

• Neuroimaging
• Radiomics
• Network Neuroscience

Background:

• Radiomics enables the creation of individualized morphological brain networks (IMBNs).
• The influence of preprocessing factors on IMBNs remains under-investigated.
• Understanding these factors is key for reliable neuroimaging analysis.

Purpose of the Study:

• To investigate how different image types and preprocessing filters affect IMBN topological organization, test-retest reliability, and disease susceptibility.
• To provide guidance on optimal preprocessing strategies for radiomics-based IMBNs.

Main Methods:

• Comparison of IMBNs from T1-weighted images and gray matter volume maps.
• Evaluation of preprocessing filters: no filter, Laplacian of Gaussian, and Wavelet filters.
• Analysis of topological organization, test-retest reliability, and disease susceptibility in three independent datasets.
• Validation using different brain parcellation schemes and application to autism spectrum disorder data.

Main Results:

• IMBNs consistently show small-world architecture and hub structures with fair-to-excellent test-retest reliability across various preprocessing methods.
• Gray matter volume-based IMBNs demonstrate superior test-retest reliability compared to T1-weighted image-based IMBNs.
• No filter and Laplacian of Gaussian filters yield higher test-retest reliability than Wavelet filters.
• Altered morphological connectivity and network topology in autism spectrum disorder were detected exclusively in gray matter volume-based IMBNs, correlating with clinical traits.

Conclusions:

• Image types and preprocessing filters significantly influence radiomics-based IMBNs.
• Gray matter volume maps and specific filters (no filter, LoG) enhance the reliability of IMBNs.
• Optimal preprocessing strategies are essential for robust findings in neuroimaging research and clinical applications, particularly in disorders like ASD.