Neurocomputational Decoding of Feigned Psychosis: Frontoparietal Cognitive Control Activation and Support Vector Machine Classification for Objective Forensic Symptom Validity Assessment in Indonesia

Abstract

Background. Detecting feigned psychiatric symptoms is a high-stakes problem in Indonesian forensic evaluation (visum et repertum psychiatricum). Current assessment relies on clinical interview and psychometric symptom validity tests, which can be manipulated.

Objective. To examine whether functional magnetic resonance imaging (fMRI) combined with supervised machine learning can separate feigned from genuine psychosis in an Indonesian forensic cohort.

Methods. In this multi-center case-control study, 90 right-handed men formed three groups of 30: malingered psychosis (MAL), genuine psychosis (PAT) and healthy controls (HC). Participants completed clinical and symptom validity assessment and a 3 T block-design symptom endorsement task. Activation (MAL > PAT) was modeled in SPM12. Support vector machine (SVM) and random forest classifiers using regional activation and connectivity features were evaluated with stratified 10-fold cross-validation.

Results. MAL > PAT showed greater activation in the dorsal anterior cingulate cortex (Z = 5.31), bilateral dorsolateral and ventrolateral prefrontal cortex, anterior insula and inferior parietal lobule (family-wise error cluster-corrected p < 0.05). A radial basis function SVM on combined features classified MAL versus PAT with accuracy 83.3% (50/60), sensitivity 80.0%, specificity 86.7% and area under the curve (AUC) 0.88; a random forest reached 81.7% (49/60). Estimates come from one cross-validation without independent validation, and MAL was defined by psychometric tests.

Conclusion. Frontoparietal and cingulo-insular activation patterns are candidate markers that separated feigned from genuine psychosis in this sample and are consistent with greater cognitive-control demand during feigned endorsement. They may become a useful adjunct to conventional evaluation only after independent validation.