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

• DOI: 10.3389/fpsyt.2023.1001085
• Publication Date: 2023-04-20T14:26:17Z
• Authors: David Popovic; Maximilian Wertz; Carolin Geisler; Joern Kaufmann; Markku Lähteenvuo; Johannes Lieslehto; Joachim Witzel; Bernhard Bogerts; Martin Walter; Peter Falkai; Nikolaos Koutsoul...; Kolja Schiltz
• Keywords: Psychiatry; Support vector machines; RC435-571; forensic psychiatry; support vector machines; 3. Good health; Neurology and psychiatry; machine learning; 5. Gender equality; pedophilia; Machine learning; child sexual abuse (CSA); Forensic psychiatry; Pedophilia; Mri; MRI

Background Child sexual abuse (CSA) has become a focal point for lawmakers, law enforcement, and mental health professionals. With high prevalence rates around the world far-reaching, often chronic, individual, societal implications, CSA its leading risk factor, pedophilia, have been well investigated. This led to wide range of clinical tools actuarial instruments diagnosis assessment regarding CSA. However, neurobiological underpinnings pedosexual behavior, specifically hands-on pedophilic offenders (PO), remain elusive. Such biomarkers PO individuals could potentially improve early detection high-risk enhance efforts prevent future Aim To use machine learning MRI data identify individuals. Methods From single-center male cohort 14 15 matched healthy control (HC) individuals, we acquired diffusion tensor imaging (anisotropy, diffusivity, fiber tracking) in literature-based regions interest (prefrontal cortex, anterior cingulate amygdala, corpus callosum). We trained linear support vector discriminate between HC using these WM microstructure data. Post hoc , investigated model decision scores with respect sociodemographic (age, education, IQ) forensic characteristics (psychopathy, deviance, violence) subpopulation. assessed specificity an external 53 Results The classifier discriminated from balanced accuracy 75.5% (sensitivity = 64.3%, 86.7%, P 5000 0.018) out-of-sample correctly 94.3%. predictive brain pattern contained bilateral fractional anisotropy diffusivity left structural prefrontal cortex-amygdala connectivity both hemispheres. was associated number previous child victims, current stance on sexuality, professionally violent reoffending. Conclusion Aberrant white matter prefronto-temporo-limbic circuit be potential correlate at reoffending Although preliminary exploratory this point, our findings highlight general MRI-based particularly patterns preventive efforts.