Back to Blog List

Signs of Mental Health in Social Media Photos?

     

Many of us are chronicling our daily lives on social media. But could those photos you’re posting hold clues to your mental health? A new study looks at posts on Instagram to see if they can identify people with depression.

mental-health-instagram

Previous research looking at social media to detect mental health conditions has focused on analyzing text. Researchers Andrew G. Reece with Harvard University and Christopher M. Danforth with the University of Vermont wanted to explore whether photographs could be used to identify and predict depression. They looked to Instagram, one of the most popular social media platforms with more than 700 million active users.

Reece and Danforth analyzed more than 43,000 photographs posted to Instagram by 166 study participants. Among the participants, 71 had a previous diagnosis of depression.

They looked at a range of characteristics of the images and related Instagram metadata, including

  • How many people were present in the photos – depression is associated with social isolation
  • Color/brightness and use of Instagram filters – research has associated darker and grayer colors with negative mood
  • Posts per user per day – a measure of user activity
  • Number of comments and likes for each post – measures of community engagement

The researchers used a variety of computational methods from machine learning and image processing to analyze the photos.

Photos posted by depressed individuals were generally bluer, darker and grayer than those posted by healthy controls. Posts by people with depression received more comments but fewer likes than posts by non-depressed people. Depressed participants were more likely to post more frequently and less likely to use filters. But when they did use filters they were more likely to use the filter that converts color to black-and-white.

Depressed participants were more likely to post images with faces, but had fewer people per photo than healthy participants. This finding, according to the study authors, “may be an oblique indicator that depressed users interact in smaller social settings, or at least choose only to share experiences of this sort on social media. This would be in accordance with previous finding that reduced social interaction is an indicator of depression.”

The researchers developed models based on the data using machine learning. The model was able to identify people with depression 70 percent of the time, significantly better than general practitioners. Previous research had found that general practitioners correctly identified people with depression 42 percent of the time.

Reece and Danforth also found that depression could be detected by looking at posts made even before the date a person was first diagnosed with depression. Looking at data on postings prior to the individual’s diagnosis of depression, the researchers’ model identified only about a third of people with depression, but was correct most of the time. General practitioners identified more cases of depression, but were more likely to misdiagnose healthy subjects as depressed.

The study authors suggest that combining analysis of photos and text (comments, captions, tags) could prove better at identifying people with depression than either on its own.

Reference

Reece, AG and Danforth, CM. Instagram photos reveal predictive markers of depression. EPJ Data Science, 2017, 6:15.

Help for Patients & Families

Learn about common mental disorders, including symptoms, risk factors and treatment options. Find answers to your questions written by leading psychiatrists, stories from people living with mental illness and links to additional resources.

Learn More

     

AnxietyBipolar DisordersSleep DisordersDepressionPatients and FamiliesHoarding DisorderEating DisordersGambling DisorderSomatic Symptom DisorderPostpartum depressionAddictionPTSD

 

Comments (0) Add a Comment

 

Add a comment

Enter the text shown in this image:*(Input is case sensitive)
* - Only comments approved by post author will be displayed here.