Eye Tracking Can Help Identify Aggressive Tendencies in Adolescents
Scientists at Privolzhsky Research Medical University (PRMU) and the Centre for Language and Brain Studies at the HSE Campus in Nizhny Novgorod have developed methods to diagnose autoaggressive and heteroaggressive tendencies in adolescents using eye-tracking technology. This will allow psychiatrists and psychologists to address this condition in young people before they may harm themselves or others. Anna Khomenko, Senior Research Fellow of the Centre for Language and Brain Studies at the HSE Campus in Nizhny Novgorod, presented the results of the project at the NeuroThursday seminar.
The project is implemented by PRMU as part of the Priority 2030 programme of the Science and Universities national project.
Autoaggression refers to aggression directed at oneself, while heteroaggression is aimed at the external world. Experts stress the importance of early and accurate diagnosis of auto- and heteroaggressive tendencies to prevent harmful actions.
Lala Kasimova
'Unfortunately, not every child chooses to seek help in coping with their emotions. It is crucial to identify the need for such assistance, which is why we propose an alternative strategy to support children facing various life challenges. This approach involves collaborative efforts between parents and a school counsellor,' explains Lala Kasimova, Doctor of Medical Science, Head of the Department of Psychiatry at PRMU.
Psychiatrists from PRMU, psycholinguists from the Centre for Language and Brain Studies at HSE in Nizhny Novgorod, law enforcement representatives, and developers from the Russian IT company Lad have collaborated to create reliable diagnostic tools.
Experts at Lad have developed eye-tracking technology and analytical algorithms. The program tracks the movement of the pupils and the duration of their focus on specific objects displayed on the screen using a webcam. In collaboration with experts in neurolinguistics, stimulus material was developed—slides containing text and images that school students are instructed to view during testing.
Anna Khomenko
'We hypothesised that if a person frequently uses the word "bullying" in their speech, for example, this word is likely to trigger a reaction when they read it. The reaction will be reflected in the movement of their gaze: the person will likely return to the word multiple times or focus on it for a longer period,' according to Anna Khomenko, Senior Research Fellow at the Centre for Language and Brain Studies, HSE Campus in Nizhny Novgorod.
To create stimuli, the neurolinguists used publicly available datasets as well as data provided by PRMU psychiatrists to study the speech profiles of individuals prone to auto- and heteroaggression. Representatives of both groups have distinct linguistic characteristics. Thus, autoaggressive individuals tend to use a higher number of first-person pronouns. Texts related to loneliness and the search for tactile contact resonate with this group. They also tend to avoid the word 'death.' Individuals with severe heteroaggression often exhibit xenophobic vocabulary. They frequently use action verbs ('do it,' 'I'm moving') and performatives ('How much longer should I put up with this, I ask you?'), and their speech is marked by accusations directed at society.
Based on the data collected, the linguists identified keywords, phrases, and expressions that informed the stimulus material for eye-tracking diagnostics. The content of this material has been reviewed for compliance with legislative norms and evaluated by psychiatrists for the appropriateness of presenting certain elements as stimuli. Some of the research stimuli were created manually, while others were generated using ChatGPT and YandexGPT3.
The scientists then began experiments, with the primary goal of testing how accurately the presence or absence of auto- or heteroaggressive tendencies can be identified using an eye tracker. The study included volunteers representing the general population, as well as individuals with confirmed cases of autoaggressive or heteroaggressive behaviour currently undergoing psychiatric treatment, either inpatient or outpatient.
So far, the scientists have completed two stages of research, with 167 participants. In the first stage, the automatic classifier correctly identified autoaggressive tendencies in 72.5% of cases and heteroaggressive tendencies with nearly 100% accuracy.
In the second stage, the scientists chose participants with less pronounced tendencies, as these individuals also often require assistance. The algorithm's accuracy then decreased to 64% for autoaggression and 23% for heteroaggression.
The researchers concluded that further fine-tuning of the algorithm is necessary to achieve more accurate results with relatively weak tendencies. Additionally, constant updating of the stimulus material is necessary, as the world around us is changing rapidly and our language reflects it. This is what the scientists continue to work on.