Gray Matter Alterations and Symptom Severity in Forensic Psychiatric Patients: An Exploratory VBM Study

ABSTRACT

Aim

The relationship between mental illness and criminal behavior is a critical social issue that remains under debate. This study aimed to investigate differences in cortical gray matter (GM) volume between mentally ill individuals institutionalized in a residence for the execution of security measures (REMS) following a violent offense—and deemed not criminally responsible due to their psychiatric condition—and a control group of healthy non‐offenders, using voxel‐based morphometry (VBM) analyses.

Method

We recruited 13 male violent offenders with psychotic disorders institutionalized in REMS and 13 healthy controls. Psychiatric symptoms were assessed using the Brief Psychiatric Rating Scale (BPRS), and psychopathy traits with the Psychopathy Checklist–Revised (PCL‐R). High‐resolution structural Magnetic Resonance scans were acquired on a 3T scanner. VBM analyses were used to identify GM volume differences between groups, and correlation analyses were performed to assess associations between brain structure and clinical measures.

Results

The experimental group showed wide variability in symptom severity and psychopathy traits. VBM analyses revealed GM volume reductions in bilateral insular cortex, in left superior temporal gyrus (STG) and right fusiform gyrus (FG) of the experimental group. GM volume in the left STG‐insula cluster significantly correlated with BPRS scores. No significant associations were found with PCL‐R scores.

Conclusion

These findings highlight structural brain alterations associated with psychotic symptomatology and should serve as a starting point for future research exploring the neural changes associated of pathological social behavior in mentally ill persons who committed violent crimes.

The relation between mental illness and criminality is a relevant social issue that has been debated over the years. In this study, we investigate differences in cortical gray matter volume in a population of mentally ill and socially dangerous persons compared to a control group of healthy nonoffender participants, using voxel‐based morphometry. We identified gray matter volume reductions in specific brain regions involved in social behavior, like saliency detection, emotional awareness, and empathy, and correlated those regions with both psychiatric symptom severity and psychopathy scores, identifying which clinical components are most strongly related to structural brain changes in mentally ill and socially dangerous individuals.

1. Introduction

The relationship between mental illness and violent behavior is a socially relevant and long‐debated issue. Severe mental disorders, such as schizophrenia and bipolar disorder, have been associated with an increased risk of violence perpetration [1]. In Italy, individuals deemed not criminally responsible due to severe mental illness may be declared socially dangerous and confined in REMS (residence for the execution of security measures), dedicated facilities for treatment and risk management based on national forensic legislation [2, 3].

The risk of future violence in mentally ill individuals depends on complex interactions between psychopathological symptoms, environmental triggers, and personality traits. This risk is notably higher in individuals with comorbid substance use or antisocial personality disorders [4, 5, 6]. Neuroimaging research has increasingly supported efforts to clarify the neurobiological underpinnings of violence in psychiatric populations [7, 8, 9, 10]. Structural imaging studies have shown reductions in gray matter (GM) volume in regions related to emotional regulation, reactive aggression, and empathy such as the frontopolar cortex, orbitofrontal cortex, insular cortex, and amygdala in persistent violent offenders [11, 12, 13, 14, 15]. Additional findings have identified GM abnormalities in the hippocampus, posterior insula, orbitofrontal cortex, and striatum among individuals accused of homicide [16, 17].

To date, neuroimaging studies have mostly investigated brain abnormalities in violent offenders, patients with schizophrenia, or individuals with psychopathy as separate groups. However, the intersection of these conditions specifically, individuals with psychotic disorders who committed violent crimes and were institutionalized in REMS following a legal assessment of social dangerousness has been largely neglected. Recent reviews have highlighted the lack of consistent subgroup definitions and dimensional approaches linking symptom severity or personality traits to brain morphology [18, 19]. To our knowledge, this is the first study applying voxel‐based morphometry (VBM) in this specific population under the Italian forensic framework.

Our investigation focused on individuals affected by psychotic disorders who committed violent crimes, were deemed not criminally responsible due to mental illness, and were confined in REMS as socially dangerous. This subgroup is of clinical and societal relevance due to their distinct legal status and high risk of recidivism. Furthermore, unlike most previous studies, we examined whether gray matter abnormalities in this population were more closely associated with psychiatric symptom severity or with psychopathic traits, using validated clinical rating scales.

By correlating GM volumes with both psychiatric symptom severity and psychopathy scores, our study adopts a dimensional approach aimed at identifying which clinical components are most strongly related to structural brain changes. This approach may improve the understanding of neurobiological correlates associated with violent behavior in mentally ill individuals and provide exploratory insights into treatment and management strategies.

2. Methods

2.1. Participants

We enrolled 13 right‐handed men (Experimental Group, EG; mean age 44 ± 7 years) in a forensic psychiatric unit (REMS, ASL RM5, Rome), each having committed violent crimes (e.g., attempted homicide, aggression, vandalism, domestic violence). All were diagnosed within the psychotic spectrum (schizophrenia, schizoaffective disorder, delusional disorder, brief psychotic disorder), frequently with comorbid Cluster B personality disorders, and deemed socially dangerous by judicial authorities with a high risk of recidivism. Their institutionalization lasted no more than 2 years.

Clinical information regarding pharmacological treatment, history and current status of substance use, psychiatric diagnosis, illness duration, and REMS residence duration was systematically collected for all patients at the time of MRI acquisition.

All participants were taking psychopharmacological therapy at the time of enrollment. Treatment consisted of second‐generation antipsychotics (risperidone, paliperidone, aripiprazole, quetiapine, clozapine), sometimes with two agents added to each other, to achieve effective stabilization of psychotic symptoms and behavioral disorders. Some participants were also taking mood‐equalizers such as lithium, sodium valproate, and pregabalin. Dosages of lithium and sodium valproate were determined based on plasma levels within the therapeutic range.

The Control Group (CG) consisted of 13 healthy, right‐handed male volunteers (mean age 38 ± 11 years) without a history of psychiatric illness, treatment, or criminal convictions, and with no MRI contraindications. There was no significant age difference between groups (independent samples t‐test: t (24) = 1.165, p = 0.127). All participants provided written informed consent after receiving a full explanation of the study procedures, which was approved by the Ethics Committee of Sapienza University of Rome (n. 727/18, CE n. 5155).

All participants in the EG underwent neurocognitive and psychiatric evaluations. Cognitive functioning was assessed using the Mini‐Mental State Examination (MMSE). Psychiatric symptom severity was evaluated using the Brief Psychiatric Rating Scale (BPRS), a clinician‐administered tool that measures common psychiatric symptoms across 18 items scored from 1 (not present) to 7 (extremely severe), with total scores ranging from 18 to 126 [20]. Psychopathy levels were assessed using the Psychopathy Checklist‐Revised (PCL‐R), a structured clinical tool scored by a trained psychiatrist. It includes 20 items rated on a 3‐point scale (0 = not applicable, 1 = partially applicable, 2 = fully applicable), with a total score range of 0 to 40. Higher scores indicate more severe psychopathic traits [21].

3. Imaging Acquisition and Analysis

A Siemens Magnetom Verio 3‐Tesla scanner was used to acquire all images. Structural scans of the brain were acquired for each participant using a T1‐weighted three‐dimensional sagittal magnetization‐prepared rapid gradient echo sequence with the following parameters: 176 slices, repetition time: 1900 ms, echo time: 2.93 ms, slice thickness: 1 mm, and an in‐plane resolution of 0.508 × 0.508 mm. We performed VBM analyses on participant T1‐weighted structural images using Computational Anatomy Toolbox (CAT12), which runs within SPM12 (www.fil.ion.ucl.ac.uk). The T1 anatomical images were manually checked for scanner artifacts and gross anatomical abnormalities. The images were then normalized using high‐dimensional Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) normalization and segmented into GM, white matter (WM), and cerebrospinal fluid (CSF). Segmented GM images were checked for quality and smoothed using an 8‐mm full‐width half‐maximum (FWHM) kernel. Total intracranial volume (TIV) was estimated for each participant and used as a covariate in the second‐level statistical test, along with age. The statistical parametric maps resulting from group comparisons were thresholded at p < 0.05, corrected for multiple comparisons at the cluster level using a false discovery rate (FDR) approach after forming clusters of adjacent voxels surviving a threshold of p < 0.001 uncorrected. The association between clinical data and average regional volume in brain areas showing a significant decrease in volume in the EG was assessed by calculating Pearson correlation coefficients.

4. Results

Participants in the EG showed a heterogeneous clinical profile regarding illness duration, antipsychotic treatment (including monotherapy or polytherapy with second‐generation antipsychotics, mood stabilizers, or lithium), and substance use history. All participants performed within the normal range on the MMSE, confirming the absence of global cognitive impairment (mean = 28; SD = 1.4). Psychiatric symptom severity, assessed with the BPRS, ranged from 35 to 75, with a mean score of 61 (SD = 9.9), indicating moderate to marked psychiatric symptoms in most individuals. Psychopathy severity, assessed using the PCL‐R, ranged from 5 to 32, with a mean score of 17.6 (SD = 7.9), reflecting a wide variability in psychopathic traits among participants. Table 1 summarizes individual demographic, forensic, and clinical characteristics of the EG, including pharmacological treatment.

TABLE 1.

Demographic, clinical, forensic, pharmacological, and neuropsychological characteristics of the experimental group.

EG ID	Age (years)	Diagnosis	Illness duration (years)	Therapy	Type of crime	Substance use	REMS duration (months)	MMSE	BPRS score	PCL‐R score
01	45	APD + BrD	26	VPA + ARI + PAL	IFV	SUD‐CE	12	27	61	26
02	53	DD	27	PAL	OJ	NO	12	29	64	8
03	42	Sc	10	QUE + PAL	IVV, OJ	SUD‐CE	22	29	66	21
04	63	Sc	44	CLZ + PAL	V	SUD‐CE	12	25	74	5
05	38	SaD	8	ARI	Va, V	SUD‐ER	1.5–2	28	35	24
06	40	BPD + BrD	6	PGB + ARI	IFV	SUD‐CE	13	30	63	16
07	47	DD	10	ARI + Li	V, IFV	NO	12	27	75	10
08	48	APD	17	VPA + OLA	V	SUD‐CE	12	26	57	23
09	40	DD	5	VPA + PAL + ARI	IFV, OJ	SUD‐CE	9	29	64	19
10	37	Ma+	15	PAL + Li	H	SUD‐CE	12	30	53	10
11	43	APD + AUD	15	VPA	IFV, OJ	SUD‐CE	21	28	57	21
12	43	APD + BrD	24	QUE + PAL	V, OJ	SUD‐CE	12	28	61	32
13	37	Sc	10	CLZ	H2	SUD‐CE	12	29	64	8

Abbreviations: AE, Antiepileptic; APD, Antisocial Personality Disorder; ARI, Aripiprazole; AUD, Alcohol Use Disorder; BPD, Borderline Personality Disorder; BrD, Brief Psychotic Disorder (with specific marked stressor); CLZ, Clozapine; DD, Delusional Disorder; H, Homicide; H2, Double homicide; IFV, Intra‐family violence (including attempted homicide); Li, Lithium; Ma+, Manic Episode with psychotic features; NO, No history of substance use; OJ, Obstruction of justice; OLA, Olanzapine; PA, Past abuse (no longer current); PAL, Paliperidone; PGB, Pregabalin; QUE, Quetiapine; SaD, Schizoaffective Disorder; Sc, Schizophrenia; SUD‐CE, Substance Use Disorder – Controlled Environment; SUD‐ER, Substance Use Disorder – Early remission; V, Violence against others (including attempted homicide); Va, Vandalism; VPA, Valproate.

All participants completed the study. Regarding structural differences between the two groups, regional clusters of GM reduction were significantly observed in the EG in specific regions, as compared with controls (Figure 1). In particular, we observed in the right hemisphere a significant GM reduction in the insular cortex (Ins) and in the fusiform gyrus (FG). In the left hemisphere, reduced GM volume was detected in the insular cortex, with the left cluster also extending to the superior temporal gyrus (Ins/STG). Table 2 lists the brain regions described above.

FIGURE 1.

The red‐to‐yellow patches show, on axial slices, the t‐statistic maps of the comparisons between GM volume in the EG and CG for p < 0.05, corrected for multiple comparisons at the cluster level using FDR and uncorrected peak p < 0.001. EG showed a significant reduction of GM volume in FG and Ins of the right hemisphere and in Ins/STG of the left hemisphere.

TABLE 2.

The table lists brain regions with reduced GM volume in EG, the hemisphere, the T‐score (p FDR corrected < 0.05), the region volume (voxels), the peak p value, and the MNI coordinates.

Region	Hemisphere	Cluster p (FDR corr)	Volume (K)	T	Peak p (unc)	x, y, z
FG	R	0.018	1164	5.9	0	38, −26, −21
				5.11	0	34, −40, −18
				4.91	0	48, −26, −24
Ins	R	0.018	1203	4.94	0	34, 20, −6
				4.48	0	32, 6, −20
				4.47	0	−36, −4, −12
Ins/STG	L	0.01	1609	4.63	0	−51, −6, 3
				4.53	0	−58, −2, −12
				4.47	0	−42, 8, −18

Note: Height threshold: T = 3.50, p = 0.001 (0.997); Extent threshold: k = 1164 voxels, p = 0.001 (0.006).

To further assess the effects of EG clinical characteristics on brain morphometry, we also performed a correlation analysis between psychiatric symptoms and regions with reduced GM volume. The clusters in Ins/STG of the left hemisphere significantly correlated with symptom severity as measured by the BPRS (r = −0.542; p = 0.028) (Figure 2). No other significant associations were detected.

FIGURE 2.

Scatterplot depicts the correlation between the BPRS score and GM volume reduction in Ins/STG of the EG.

5. Discussion

In this VBM study, we identified GM volume reductions in a small group of mentally ill individuals institutionalized in a REMS psychiatric unit due to social dangerousness and a history of violent crime. Compared to healthy controls, these subjects showed significant GM volume loss primarily in specific brain regions as Ins and FG of the right hemisphere, and in Ins/STG of the left hemisphere. Furthermore, within the EG, GM volume in the Ins/STG cluster negatively correlated with psychotic symptom severity as measured by the BPRS. These findings expand upon our prior resting‐state fMRI study on the same population [10], where we observed aberrant functional connectivity within brain networks involved in morality, salience attribution, and reward processing, including the orbitofrontal cortex, nucleus accumbens, amygdala, and cingulate cortices. The present study adds to this by revealing structural abnormalities in regions anatomically and functionally linked to these same networks. In particular, the insula's role in integrating interoceptive and emotional information suggests a disruption in emotional awareness and empathy in the EG.

Our findings align with growing evidence of GM reductions in frontotemporal and limbic regions among violent individuals with psychiatric disorders. A recent voxel‐based meta‐analysis by Wang et al. (2023) [22] confirmed consistent reductions in areas such as the bilateral insula, STG, and orbitofrontal cortex, which are crucial for emotional regulation and impulse control. Numerous structural MRI and VBM studies have similarly reported reduced volume in the STG and insula in individuals with psychosis and those with antisocial traits including psychopathy and callous‐unemotional features supporting the involvement of these regions in the overlap between psychotic and antisocial dimensions [11, 23, 24, 25].

Complementing these findings, a large‐scale study by Haukvik et al. [18] using normative modeling identified heterogeneous morphometric deviations particularly in the lingual gyrus, collateral transverse sulcus, and cerebellum in individuals with schizophrenia spectrum disorders and violent histories. Interestingly, no strong associations with psychopathy traits emerged, possibly due to methodological differences focused on deviation scores rather than dimensional symptom correlations.

In contrast to our findings, a recent VBM study by Chou et al. (2022) [26] reported increased GM volume in regions including the bilateral middle frontal gyrus, anterior cingulate cortex, and STG in both affective and predatory violent offenders compared to controls. These differences may reflect distinct sample characteristics, as their study included nonpsychotic violent offenders, whereas our cohort consisted of individuals with psychosis, deemed legally nonresponsible and institutionalized in REMS due to high recidivism risk. Moreover, the volumetric increases observed by Chou et al. have been interpreted as potential compensatory neural mechanisms or trait‐based differences in cognitive control and emotional processing, whereas our reductions in empathy‐ and salience‐related areas are more consistent with prior studies on psychosis‐related violence.

Our study advances this literature in three ways. First, it focuses on a highly specific and under‐investigated population individuals institutionalized in REMS with a formal legal classification of high social dangerousness thereby isolating a group with both psychiatric illness and forensic relevance. Second, unlike many previous studies based solely on group comparisons, we adopt a dimensional approach and correlate GM volume with both psychotic and psychopathic traits, thus aiming to explore the neural correlates potentially underlying distinct psychological dimensions. Finally, we observed a pattern of structural deficits (especially in Ins/STG) that appears consistent with the clinical features of our sample: severe psychopathology, emotional dysregulation, and a history of interpersonal violence.

The most salient result of our study was the regional volumetric reduction in brain areas of the EG implicated in empathy, emotional awareness, and regulation, namely the bilateral insula. The insular cortex is a critical site where bodily sensations, autonomic control, and afferents from brain regions implicated in emotional processing converge [27, 28]. The insula represents the brain hub where incoming external sensory information is integrated with the internal physiological response related to emotional state [29, 30]. Early theories of emotions emphasized the link between interoception and emotions by arguing that emotions are evoked by the perceptions of physical responses of the body and cannot exist without the experience of bodily feelings [27]. A striking example of a deficit in sensory‐emotional integration is that of pain asymbolia, in which patients suffering from an insular lesion can recognize pain, but lack an appropriate emotional response and do not attribute a negative valence to this usually adverse experience [31]. Recent findings have demonstrated that the right insula contributes to emotional feeling states originating in representations of visceral arousal [32]. GM volume reduction in the bilateral insula has also been reported in offenders with psychopathic traits [12], suggesting that similar neural alterations underlie criminal psychopathy and variation in antisocial behavior. Differential roles of the right and left insula in representing sympathetic and parasympathetic nervous system activity have been proposed by Craig et al. in a homeostatic neuroanatomical model of emotional asymmetry, suggesting that the right insula predominantly represents sympathetic activity, while the left insula reflects parasympathetic modulation [33]. The bilateral insular finding in our study may thus reflect the involvement of both sympathetic and parasympathetic representations in the experience of empathy and emotional regulation in the EG. Insular cortex, a key node of the salience network alongside the dorsal anterior cingulate cortex, plays a central role in integrating bodily states with contextual and emotional stimuli [27, 34, 35]. Saliency detection can be conceptualized into two general mechanisms. The first is a fast, automatic, bottom‐up “primitive” mechanism for filtering stimuli based on their perceptual features; the second is a higher‐order system for focusing the “spotlight of attention” and enhancing access to resources needed for goal‐directed behavior [36]. The misattribution of salience to external and internal stimuli in individuals with mental illness is a core feature of the disorder and may explain the genesis of psychotic symptoms [37]. Indeed, salience dysfunction has been linked to violence, aggression, psychopathy, criminal behavior, and recidivism [38, 39]. The insular cortex has also been associated with the integration of sensory phenomena and mediates temporally defined auditory/visual interaction at an early stage of cortical processing [40, 41]. The reduced GM volume in auditory areas like the STG or the FG (implicated in facial recognition and the identification of facial expressions [42]) suggests that multiple brain regions beyond the insula may be involved in altered empathic processing and socio‐emotional functions, which could be related to difficulties in emotional regulation and, in some cases, to aggressive behavior in the EG.

The STG is a site of multisensory integration [43]. It is implicated in spoken word recognition and in visual analysis of social information conveyed by gaze and body movement [44, 45]. A progressive GM reduction in the STG has been found to precede the first expression of florid psychosis [46]. Here, we found a significant correlation between GM volume reduction in the Ins/STG cluster in the left hemisphere and psychotic symptom severity, as scored by the BPRS. These findings are in line with previous VBM studies that have demonstrated that high‐risk subjects who later developed psychosis exhibited progressive GM reduction in left temporal lobe regions, particularly the STG. Furthermore, the STG was found to be reduced in criminals with diminished responsibility due to personality disorders [23] pointing to frontotemporal dysfunction in patients with psychopathy [47]. Volume reduction in these regions has been found to correlate with auditory hallucinations or thought disorders [48, 49] and with the severity of antisocial behavior of offenders [24]. Changes in insular and STG morphometry and the consistent correlation between GM volume and psychotic symptom severity in the EG suggest that the left insula and the STG contributed to psychotic symptom manifestation in this group through their significant interaction. However, it remains to be determined whether these morphometric changes are a cause or consequence of social dangerousness.

Together with STG, facial and social perception are elaborated by regions in the FG, which are mainly involved in facial expression recognition [50], especially when viewing negative faces [51]. Recent studies found that viewing violent scenes elicited widespread activation, including the FG and STG, both in violent and nonviolent subjects [52]. Furthermore, GM reduction was found in brain areas involved in emotion processes, such as limbic brain regions (insula and parahippocampal) [53]. GM reduction in the FG, STG, temporal pole, and insula has been reported in violent patients with schizophrenia, suggesting that deficits in emotion‐related regions may predispose individuals to violent behaviors [54]. Specifically, Gou and colleagues [54], speculated that GM reduction in the FG might predispose individuals to a persistent risk for violence in schizophrenia patients due to deficits in emotion perception, especially for negative faces.

Our study has several limitations. First, we tested socially dangerous men because the REMS was male only; thus, generalizations to women cannot be made. Second, the EG was composed of socially dangerous persons, convicted of crimes, with a wide heterogeneity of psychiatric comorbidities, familial and environmental factors that may have affected our results, thus limiting generalizability to other clinical populations. Third, the absence of a nonviolent psychiatric control group limits the ability to isolate the effects of violent behavior per se; however, our dimensional analysis of psychotic symptoms and psychopathy traits partially addressed this issue. Finally, the small sample size limits the generalizability of the findings.

6. Conclusions

The relation between psychiatric illness and criminality represents a relevant social issue and a topic of intense debate involving psychiatrists, lawyers, and neuroscientists. Improved technology has made high‐quality neuroimaging data more accessible and exploitable for scientific research and in clinical practice, and it is widely accepted that neuroscience variables can be used to improve existing behavioral prediction methods in the evaluation of recidivism.

Our findings may serve as a starting point for future research exploring the neural correlates of pathological social behavior in mentally ill persons who committed violent crimes and could contribute to a better understanding of the neurobiological mechanisms associated with violence in psychosis. These structural observations may also inform the development of neurobiologically grounded frameworks for forensic evaluation, potentially supporting more accurate and individualized approaches to rehabilitation.

Author Contributions

All authors contributed to the study conception and design. Methodology Maddalena Boccia, Gaia Cartocci; Formal analysis and investigation Gaia Cartocci, Pieritalo Maria Pompili, Maddalena Boccia; Writing original draft preparation Gaia Cartocci, Maddalena Boccia, Pieritalo Maria Pompili, Francesca Caramia; Writing review and editing Gaia Cartocci, Pieritalo Maria Pompili, Maria Giulia Tullo, Francesca Caramia; Supervision Stefano Ferracuti, Vittorio Fineschi, Paola Frati, Marco Fiorelli. The authors certify that they have no affiliations with or involvement in any of the following organizations: National Institutes of Health (NIH); Wellcome Trust; Howard Hughes Medical Institute (HHMI); and other (s) or entity with any financial interest, or nonfinancial interest in industry supporting the project.

Ethics Statement

The present study was approved by the local ethics committee of Sapienza University of Rome (727/18, CE n.5155).

Consent

All participants gave written consent for publication. Participants' anonymity was ensured, also because they were declared socially dangerous, and they were detained in special residences for the execution of security measures after the decision of the judicial authority.

Conflicts of Interest

The authors declare no conflicts of interest.

Cartocci G., Boccia M., Pompili P. M., et al., “Gray Matter Alterations and Symptom Severity in Forensic Psychiatric Patients: An Exploratory VBM Study,” Neuropsychopharmacology Reports 45, no. 4 (2025): e70070, 10.1002/npr2.70070.

Data Availability Statement

The data are not publicly available due to legal restrictions e.g., their containing information that could compromise the privacy of research participants.