Virtual Reality and Speech Analysis for the Assessment of Impulsivity and Decision-making

INTRODUCTION Impulsivity is defined as the tendency to act without prior reflection. It implies poor self-control of behavior, leading to quick reactions without regard for consequences. Impulsivity is strongly associated with suicidal behavior and may define a particular suicidal phenotype in which action predominates over ideation.

The dimensional model of impulsivity considers three factors: decision-making, action, and personality traits. The first two factors are the behavioral domains of impulsivity, assessed with neuropsychological tests, while impulsivity as a personality trait is assessed with personality questionnaires.

Among these factors, the one that has been most strongly related to suicidal behavior is decision-making. Decision-making is a complex cognitive function in which learning, previous experience, and sensitivity to feedback interact. It can be intuitive or reflexive, involving different brain structures: intuitive processes are mediated by the amygdala, ventral striatum, and orbitofrontal cortex, while reflexive processes are mediated by the dorsolateral prefrontal cortex, anterior cingulate, and posterior parietal lobe. People with suicidal behavior have difficulty making decisions in risky situations, and their learning to make these decisions does not benefit from feedback. They tend to decide more emotionally than rationally. Studies have found that more than half of all suicide attempts are impulsive. Furthermore, increased impulsivity and aggression have been associated with death by suicide.

Self-administered questionnaires are the most common approach to assess impulsivity. Despite its usefulness, self-report has significant limitations: the cultural influence of the questionnaires hinders their generalizability. Furthermore, impulsive people may find it challenging to describe their personality traits and may thus report unreliably. Computerized neuropsychological tests overcome some of these limitations, as the cultural background does not influence them, and they do not rely on the capacity for introspection and self-awareness, but they are not without limitations, mainly the time it takes to administer and score the tasks. In addition, the situations and stimuli presented to the persons assessed are not very ecological, i.e., they do not usually have a translation to real life, which may compromise their validity. Virtual reality (VR) may overcome these limitations. There is empirical evidence showing that the neural mechanisms that people experience when immersed in a VR environment are similar to those of real-life. Serious VR games have become an innovative, engaging, and customizable way to provide healthcare. VR has been applied in mental health interventions and the field of neuropsychological assessment with good results.

Speech has been increasingly studied for mental health assessment and particularly as a biomarker for mental health. Speech samples can be collected easily and consistently from patients and can help to solve the problem of limited or intermittent assessments due to cost and time to see a clinician while providing ecologically relevant information that is generally not available to clinicians. Combined with machine learning techniques, speech features (textual features and/or acoustic features) are used for automatic assessment of patients and diagnosis of psychiatric disorders.

This study presents the VR assessment tool "Spheres \& Shield Maze Task", a virtual maze that participants must go through while overcoming several obstacles, and which can decrease administration time and provides an immersive experience that can approximate real-life conditions, thus increasing ecological validity. The VR tool has been previously evaluated in a non-clinical setting. This work aims to validate the Spheres \& Shield Maze Task by comparing it with traditional measures of impulsivity, decision-making, and personality traits in psychiatric patients with or without a history of suicidal thoughts and behaviors and in healthy controls. The secondary objective is to validate the implementation of speech analysis in clinical settings and examine how speech correlates to impulsivity, decision-making, impulsive personality traits, and jumping to conclusions.

The hypothesis is that both the VR assessment tool and speech analysis will be reliable and acceptable by patients, potentially improving the simultaneous assessment of impulsivity and decision making.

METHODS \& ANALYSIS This is a cross-sectional study that will be carried out in the Department of Psychiatry of the University Hospital Fundación Jiménez Díaz. Data will be collected over an estimated period of one year.

The estimated sample size is 300 participants, 100 per branch. Psychiatric patients will be recruited by their attending psychiatrist during regular visits. Healthy volunteers will be recruited following the snowball methodology generated from mental health center workers, trainees, and medicine and psychology students.

After signing the informed consent, participants will be scheduled for the assessment session, which will last approximately one hour. They will then be taken to another room and administered the VR task that is recorded by the dual microphones in the Vive Pro Eye VR headset. Trained psychologists will administer the questionnaires and assist the participant during the VR task.

The following assessments will be carried out: 1) collection of socio-demographic and clinical data, 2) self-reported assessment of impulsive personality traits, 3) computerized neuropsychological tests, 4) VR-based impulsivity assessment, 5) assessment of jumping to conclusions, 6) recording of spontaneous speech during both the assessment session and VR task, and 7) satisfaction surveys.

An ad hoc questionnaire will be developed to record sociodemographic variables of interest, including gender, age, marital status, employment status and educational level. History of suicidal behaviour will also be assessed using the Columbia Suicide Severity Rating Scale.

Additional questionnaires will be the patients' health questionnaire-9 (PHQ-9), the PANSS scale -only in case of psychosis-, and the General Anxiety Disorder-7 (GAD-7) questionnaire. Participants will also be inquired about substance use -including alcohol, tobacco, cannabis, cocaine, hallucinogens, and other drugs.

Impulsive personality traits will be measured through the Barratt Impulsivity Scale (BIS-11), and the Urgency, Premeditation, Perseverance, Sensation Seeking, Positive Urgency, Impulsive Behavior Scale (UPPS).

Computerized neuropsychological tests will be the Iowa Gambling Task (IGT), the Continous Performance Test (CPT), the Stop signal Task, and the Go/no-go Task.

For speech analysis, the Emotional Affective Questions Good News Question and the Sadness Characteristic Question will be used. Additionally, the patient is asked to respond for two minutes and the interviewing psychologist will refrain from speaking during this period and timestamp the recording. This question is designed to elicit an emotional response related to positive and negative emotions.

The VR environment, called Spheres \& Shield Maze Task, has been developed by the Laboratory of Immersive Neurotechnologies (LabLENI) of the Universitat Politècnica de València. To implement the virtual environment, a VIVE pro eye VR headset, a VNET INTEL I7 computer and a Shimmer GSR (galvanic skin response) kit will be used. The game consists of a maze that the user must solve by moving through space. The movement is done by combining a touchpad with front-facing orientation.

Different hazards will present throughout the maze, with time penalty, energy loss, and certain consequences, such as narrow walkways from which the player can fall, lightning clouds that strike in intervals, flying insects, or locked rooms. The user can also make use of a shield that allows moving through lightning and insects without time penalty at the cost of being unable to gather green orbs. Hazards may also be avoided completely by choosing another path through the labyrinth, however completion of the maze requires traversal of a minimum number of hazards.

The screen displays an indicator of the energy which increases by gathering green orbs and decreases with movement. If the energy decreases, the movement is slowed down. These consequences may influence the user's decision-making. Energy can be recharged by consuming green spheres to be found along the route.

The maze will be scored as the time taken to find the exit. Additionally, data will be collected on behavioral and psychophysiological measures during the execution of the task, including distance travelled, number of stops, trajectory, decision making at bifurcations, total energy obtained, number of spheres collected, use of the Shield, and use of the break room.

At the end of the session, participants will complete a quantitative and a qualitative satisfaction survey, which have been used previously.

Statistical analyses will be carried out using SPSS 24.0, R, Matlab and Python statistical software. External consistency of the VR task will be assessed using traditional neuropsychological measures as Gold Standard, and internal consistency will be assessed using Cronbach's alpha and intraclass correlation coefficient. To explore the differences between groups (suicidal patients, non-suicidal patients, and healthy controls), the mean scores obtained in the variables collected will be compared using ANOVA test or its non-parametric equivalent if the distribution is not normal. Audio recordings will be analyzed in OpenSMILE, an open-source offline software for audio analysis. Machine learning techniques will then be used for feature selection and examining the relation between a participant's speech, impulsivity, suicidality, and mental health questionnaires. Additionally, analysis of recorded speech or utterance before and during the VR task will be used to predict satisfaction of the participant. Descriptive statistics from the quantitative satisfaction survey and content analysis for the qualitative satisfaction survey will also be provided. All tests will be two-tailed with statistical significance stablished at p<0.005 and 95% confidence intervals.

DISCUSSION It is expected that the VR-based assessment to be comparable to the traditional self-report and neuropsychological assessment and to be better accepted by the participants. This study will also contribute to increasing knowledge about the association between impulsivity and suicidal behavior and the use of speech analysis in a clinical setting.

This study will result in a series of manuscripts that will be submitted to peer-reviewed journals for publication. Both significant and non-significant results will be reported. The findings will also be presented in the form of oral communications and symposia at national and international psychiatric conferences.