Hidden Stories of Medical Conditions


How Visualization and AI can help discover hidden patient stories from online data?

Visualisation: Launch demo

Med-DL

Every single day millions of patients globally visit their doctor and discuss the medical conditions they experience. Most often, the doctors will focus on the physical aspects of patients’ conditions, such as fever, pain, or high cholesterol, and prescribe medications and remedies. However, in every single case, the patients experience also other aspects of their conditions, such as low motivation at work, desire to isolate from company, and feelings of depression or even shame (i.e., psychological and social aspects). Only some among such issues the patients will share with the doctors and even those that they share, the doctors might consider less relevant than physical and so overhear them. At the same time, medical researchers are suggesting already for some decades that the psychological and social aspects of medical conditions might be key in actually helping the patients.

Now imagine if we could collect all these untold patient stories to communicate experiences of any given medical condition? The researchers from Nokia Bell Labs did exactly this in their recently published study in the IEEE Visualization Conference (InfoVis). Upon applying the state-of-the-art AI techniques for natural language processing, they extracted mentions of symptoms, conditions, feelings, and body parts from online discussions by patients on Reddit. To discover patient stories, they worked with subreddits (Reddit’s communities) that are focused on specific conditions such as depression, anxiety, rheumatoid arthritis, and psoriasis. These are online spaces where people suffering from those conditions come to share theirs and learn from the experiences of others, and, most importantly, to support each other.

Upon processing half a million anonymized Reddit posts from such communities, the Bell Labs researchers developed a visualization tool, which enables exploring the uncovered collective experiences of different conditions. They have also evaluated their visualization in a user study with medical experts, whose comments had helped to shape it. In a later user study with the general public, the researchers found their visualisation more effective compared to textual stories in communicating each of the three aspects of medical conditions (i.e., physical, social, psychological).

Med-DL
Med-DL

Publications

  • Humane Visual AI: Telling the Stories Behind a Medical Condition. IEEE VIS 2020. PDF
  • Extracting Medical Entities from Social Media. CHIL 2020 PDF