The Weekly Mind Reader: Facebook Images Help Predict Suicide Risk

As part of The Journal of Clinical Psychiatry’s ongoing Spotlight on Suicide series, a new study tackled the concept of suicide prediction using a detailed analysis of images posted on Facebook. 

Unique Features

The aim of the study was to assess an individual’s suicide risk based on the images they share on the world’s most popular social media platform. Several studies have looked at predicting suicidal ideation using other types of artificial intelligence models to scan social media sites but for the most part they’ve focused on reviewing the text portion of posts versus their visual imagery. 

Researchers reviewed a dataset of 177,220 images from 841 Facebook members examining a unique set of three features:

Overall Images: This cluster included tasks related to image brightness, sentiment, and content. It differentiated between bright and dark settings, positive and negative sentiments, and content such as the presence of humans, animals, and non-living objects.

Single Person: In images containing a single person, the study focused on identifying whether it was a selfie or not, the emotional state of the person (e.g., happy or sad), and their developmental stage (e.g., child, adult, elderly).

Multiple People: For images with more than one person, the analysis included determining if it was a selfie, the emotional state of the people ( e.g., happy or sad), and the nature of their relationships (e.g., romantic couples, families, friends, or work colleagues).

Overall Outcomes

Results indicated that the images were highly predictive of suicidal ideation. In fact, with a 72 percent rate of accuracy, the method far surpassed traditional deep learning models. The effect size, measured by Cohen’s d, came in at at 0.82, indicating a meaningful impact for differentiating between high and low suicide risk individuals. 

When compared with established models like ResNet, this new model also showed superior predictive capabilities, with the detailed analysis revealing 11 out of 24 features. This demonstrated clear variations between high-risk and low-risk groups. Particularly, 8 of these features were found to be strong suicidal tendency predictors, emphasizing aspects like negative emotions and the relational content in images. Furthermore, the model’s predictions closely match real-world outcomes, as shown by near-perfect calibration scores. 

According to the researchers, the model proved itself as a valuable source of information that will help them develop better monitoring tools. They also concluded that the model would be more accessible and easier for clinicians and researchers to use compared to many other current methods. 

The authors did acknowledge certain limitations in their work, such as reliance on self-reporting and the focus on a single social media platform. Nevertheless, they believe they’ve hit on a good way of identifying the relationship between non-verbal online behaviors and suicide prevention. Overall, AI seems capable of making a real contribution to the field of suicidology and digital mental health.

IN OTHER PSYCHIATRY AND NEUROLOGY NEWS

• This study provides the first evidence of the long-term safety of a 14-day treatment course with zuranolone.
• Most patients with schizophrenia have significantly poor sleep quality that can have a deleterious impact on their daytime functioning.
• A Letter to the Editor cautions clinicians to weigh the risks and benefits of deprescribing benzodiazepines.
• Results from the SHORELINE study responders to zuranolone experienced improvements in depressive symptoms with initial and repeat treatment courses.
• A new study from the UK Biobank looked at 15 modifiable risk factors associated with young-onset dementia. Twelve were harmful. Three were protective. 

NEW AT CME INSTITUTE

Click to earn free accredited CME credit.

Optimizing Mental Health for Women: Recognizing and Treating Mood Disorders Throughout the Lifespan

Reviewing Non-Dopaminergic Mechanisms for Positive and Negative Schizophrenia Symptom Management

Untangling the Complexity of Alzheimer Disease With Effective Diagnostic Tools and Novel Treatment Options