AI in healthcare: What I learned from using diagnosis apps thanks to my constant headaches

I am a certified Medical Device Usability Expert (TÜV) and the topic I am really into is artificial intelligence (AI) in healthcare. Machine learning has become an integral part of applications and systems in health.

Examples of machine learning in healthcare applications and systems:

• MIT researchers developed a neural-network model that discovers speech patterns in peoples’ conversations or interviews indicative of depression to detect those in their early stages.
• Google’s DeepMind Health combines machine learning and neuroscience to build learning algorithms into neural networks that mimic the human brain.
• The FDA approved a system that analyzes CT scans and notifies healthcare providers of potential strokes in patients.

I had a lot of sleepless nights lately, recurring headaches, and problems with my skin. Instead of spending hours in a doctor’s office waiting area, I decided to address it as a human factors researcher and checked the web to see if AI could help me.

 

Four diagnosis apps - one winner

I tried WebMD, Babylon Health, Symptom-Checker and Ada. While WebMD, Ada and Babylon Health are apps, Symptom-Checker is a web-based tool from NetDoktor. Symptom-Checker shocked me the most. It diagnosed, after a few short questions, that I have Lewy-Body-dementia or that I could be in menopause, not even asking how old I am (I am 26 by the way). “Luckily” the tool recommends also seeing a doctor. Probably not a bad idea given this weird diagnosis. WebMD and Babylon Health were a bit more precise in asking questions. What really surprised me was Ada. As a medical device usability expert, I really need to say that the developers did a good job! On top of a clean visual design, Ada is very easy to use. The app asks you simple questions and then compares your answers to similar cases to find a possible explanation for the symptoms. In the end you get a suggestion what you should do according to the results, e.g. if you should see a doctor immediately.

 

 

Great usability and a personal addressing create trust and comfort

What I like about Ada is that it feels very personal. Despite all data protection and security concerns I have the feeling that I can trust it. Creating this feeling of individual care and trust is an important factor in healthcare. Ada uses active and normal (not highly professional) language like "I can help you find out what's going on" or "We started an assessment about your headaches. Would you like us to continue?" This professional experience is probably linked to the fact that Ada was first developed for physicians to support the diagnosis and treatment process. Later it was rolled out for patient use as well. With this fact in mind, the app and results feel even more reliable and trustworthy.

The questions it asks are simple, and it uses multiple choice answers, which saves you from cumbersome typing (particularly helpful when you have a headache).

 

The app also helps you select the right answer by using pictures to help explain what they mean and you can provide feedback at any time during the assessment, so developers constantly collect feedback to improve the app.

 

Summary

At the end of the process, I got the feeling that the suggested diagnoses actually matched my health condition – stress and tension headaches. Ada describes in detail what each of the possible causes mean and gives you a suggestion to see a doctor, call an emergency or if you can, manage it at home. Luckily, I could solve my diagnoses at home with some relaxation time, sleep and Yoga.

My professional first impression is that Ada is a very well-developed app. I am sure that less tech-savvy people like my grandparents could use Ada. It provides good guidance through the assessment process. For people like my grandparents, living in a small village with less infrastructure around, Ada could help to get a first result when not feeling well.

 

Will AI replace the doctor in the future?

I don't think so, especially from the human perspective. The healing power of human interaction, i.e. with a caring doctor or nurse, cannot be replaced by an algorithm. The AI needs doctors as much as doctors might need AI. It needs to be fed with real-life knowledge to become better. The challenge will be to find a way for both worlds to complement each other to improve treatments, diagnosis, and other sectors related to healthcare.