Patterns of risk-Using machine learning and structural neuroimaging to identify pedophilic offenders

Area of Science:

• Neuroscience
• Machine Learning
• Forensic Psychology

Background:

• Child sexual abuse (CSA) is a significant global issue with severe individual and societal consequences.
• Pedophilia is a leading risk factor for CSA, and understanding its neurobiological underpinnings is crucial.
• Existing diagnostic and risk assessment tools for CSA lack neurobiological biomarkers for pedophilic offenders (PO).

Purpose of the Study:

• To utilize machine learning (ML) and magnetic resonance imaging (MRI) data to identify neurobiological markers in PO individuals.
• To explore the potential of these biomarkers for improving early detection and prevention of CSA.

Main Methods:

• Diffusion tensor imaging (DTI) data were acquired from a cohort of 14 male PO individuals and 15 healthy controls (HC).
• White matter (WM) microstructure data from key brain regions (prefrontal cortex, anterior cingulate cortex, amygdala, corpus callosum) were analyzed.
• A linear support vector machine (SVM) was trained to discriminate between PO and HC groups; model performance was validated externally.

Main Results:

• The ML classifier distinguished PO from HC individuals with 75.5% balanced accuracy and 94.3% out-of-sample specificity for HC identification.
• The predictive brain pattern involved WM microstructure in the anterior cingulate cortex, left amygdala, and prefrontal cortex-amygdala connectivity.
• This pattern correlated with factors such as previous victim count, current sexual attitudes, and assessed risk of reoffending.

Conclusions:

• Aberrant WM microstructure in the prefronto-temporo-limbic circuit may serve as a neurobiological correlate for high-risk PO individuals.
• MRI-based WM microstructure patterns show promise as biomarkers for CSA risk assessment.
• These findings support the potential of neuroimaging biomarkers for enhancing CSA prevention strategies.