By: Julia Canick
In a world that places a high premium on happiness, the prospect of coping with a mental illness as debilitating as depression can be frightening. Scarier still, general practitioners diagnose only about half of major depressive disorder (MDD) cases (1). In an effort to raise this unacceptably low rate of success, Andrew Reece and Christopher Danforth (of Harvard University and University of Vermont, respectively) turned their attention to an unlikely source of information: social media.
In a recent study, the two researchers combined different computational methods, including color analysis, metadata components, and face detection technology, to analyze 43,950 Instagram photos from 166 participants (2). While previous predictive screening methods had successfully analyzed online content to detect some health conditions, they had all relied on text, not visual posts. This study was different: the analysis identified hallmarks of depression not through the content itself, but through a few key qualities about what was posted.
Some of the results were not shocking: Reece and Danforth found that depressed individuals were more likely to post photos that were bluer, grayer, and darker than those posted by their non-depressed counterparts (2). Furthermore, depressed posters were more likely to use a black-and-white filter (like Inkwell) on their photos, while people not afflicted with the disorder were more likely to use filters that gave their photos a warmer tone (like Valencia) (3). This is certainly a reasonable finding; previous studies had pointed to a general perception that darker and grayer colors were associated with a negative mood (4).
Another finding was more unforeseen: depressed Instagrammers were more likely than non-depressed ones to post photos with faces—with the caveat that they had, on average, fewer faces per photo (2). Depression is strongly associated with lowered social activity, so the higher frequency of posts that featured faces is counterintuitive (5).
These findings are fascinating on their own, but the most astounding part is, perhaps, the marked increase in the percentage of successful depression diagnoses made: the computer model correctly identified 70% of the cases of depression in the study, a marked increase from the 47.3% success rate of general practitioners (1,2). The system was even able to recognize markers of depression before the date of the first diagnosis, an advance that suggests that further refinement of this algorithm may lead to better preventative measures for major depressive disorder (2).
The implications for the success of a photo analysis model like this one are staggering. The generation of information using entirely quantitative means, instead of human participation, eliminates subjective opinion, and, with it, the natural propensity for observer bias. Furthermore, the speed and automation of the diagnosis process can accelerate the search for treatment, and could give rise to future programs where doctors would, with patients’ permission, receive notifications if their patients’ Instagram posts raise red flags.
This data model isn’t perfect—for example, a major concern is that patients who are officially diagnosed with MDD may identify with the label, and therefore post content in accordance with their self-image (6). This presents the potential issue of patients perpetuating stereotypes associated with their diagnoses because they tell themselves they “must act” in a manner in keeping with the general image of MDD, instead of taking steps to break the stigma accompanying their illnesses. However, the benefits of this algorithm far outweigh the possible downfalls; Instagram garners nearly 100 million posts per day, and the utilization of these photographs as data points for analysis of mental health would be revolutionary for the advancements of both diagnostic tools and therapeutic accessibility (7)
Gray days often call for gray photos—and, now, there’s a way to harness that observation to make MDD diagnosis and treatment more accurate and accessible for those who are struggling.
Julia Canick ’18 is a senior in Adams House concentrating in Molecular and Cellular Biology.
 Mitchell, A. J. et al. The Lancet. 2009, 374(9690), 609-619.
 Reece, A. G.; Danforth, C. M. EPJ Data Science. 2017, 6(1), 15.
 Brown, J. When You’re Blue, So Are Your Instagram Photos. EurekAlert! [Online], Aug. 7, 2017. https://www. eurekalert.org/pub_releases/2017-08/ uov-wyb072717.php (accessed Sept. 24, 2017).
 Carruthers, H. R. et al. BMC Med. Res. Methodol. 2010, 10(1), 12.
 Bruce, M. L.; Hoff, R. A. Soc. Psychiatry Psychiatr. Epidemiol. 1994, 29(4), 165- 171.
 Cornford, C. S. et al. Fam. Pract. 2007, 24(4), 358-364.
 Instagram. https://instagram-press.com (accessed Sept. 24, 2017).