Volumetric Integrated Classification Index: An Integrated Voxel-Based Morphometry and Machine Learning Interpretable...

Area of Science:

• Neuroimaging
• Machine Learning
• Genetics

Background:

• Post-traumatic stress disorder (PTSD) is a complex mental health condition with significant impacts on quality of life.
• Precise neurobiological markers for PTSD remain challenging to identify despite extensive research.

Purpose of the Study:

• To investigate structural brain changes in PTSD patients using VBM analysis and machine learning.
• To explore the diagnostic potential of machine learning models and a novel interpretability tool (VICI).
• To examine the association between PTSD risk genes and neuroimaging findings.

Main Methods:

• Voxel-based morphometry (VBM) analysis on brain MRI data from PTSD patients and healthy controls.
• Application of machine learning algorithms (SVM, RF, LR) for PTSD classification.
• Development and application of VICI with SHAP analysis for model interpretability and gene expression analysis.

Main Results:

• Random Forest (RF) demonstrated high accuracy in classifying PTSD patients.
• Significant structural brain abnormalities were observed in prefrontal areas of PTSD patients.
• The VICI showed comparable classification efficacy to RF and identified associations with PTSD risk genes.

Conclusions:

• VBM analysis and machine learning show promise for PTSD diagnosis and prognosis.
• The VICI offers a potential simplified diagnostic tool and enhances model interpretability.
• Findings implicate synaptic integrity and neural development in PTSD pathophysiology.