Health Chatbots – Can they help identify skin lesions?

Health Chatbots

Health Chatbots and Artificial Intelligence (AI) in healthcare has been in the news lately. Everyone is talking about it. It should not be overseen that internet search engines like Baidu, Google, Bing, Yahoo and others have used AI since the inception of search engines and have been getting better at finding the correct information and they are also using image recognition to help find answers on skin diseases.

Are health chatbots really better than internet search engines? We took a closer look using a specific disease category; the body’s largest and fastest-growing organ: the skin (which includes skin, hair and nails). It is estimated that over 20% of doctor visits are regarding the skin [1], therefore health chatbots should provide good answers about any skin concern.

Click here to try out our FREE Skin Image Search

How many skin disease visits are there every year?

A person has on average 1.6 skin diseases a year, the most common being skin infections and eczema (dermatitis) [2]. Nearly 20% of all family doctor visits are skin concerns.Skin diseases are hard to diagnose by non-dermatologists. It is also know that a dermatologist is nearly twice as good at diagnosing skin diseases compared to a family doctor or internist, in regards to the 20 most common skin diseases [3]. It is estimated that at least 80% of people search their medical symptoms on the internet and compare their skin concern with images on the internet, and in many cases they are not correctly labelled. Most people that have a skin concern (69%), self-treat by buying prescription free medication from a pharmacy. Around 14% seek further medical advice, from a local doctor or nurse  [1]. Another problem is supply and demand, there are not enough dermatologists around, for example in the UK the dermatologist patient ratio is 1: 150 000, in the USA 1:32 000, in Africa about 1: 8 million and in China 1: 100 000 [4].

Click here to try out our FREE Skin Image Search

Why can’t I see a dermatologist today?

If there is a dermatologist in your area, you can go and see one, but it can take some time. In the US the average time to see a dermatologist is around 30 days, in Sweden it can take up to 90 days and in Ireland and the UK, 12-16 months is not uncommon. However you can get an appointment quickly if it is deemed a medical emergency, like skin cancer. Referrals from family doctors help dermatologist to prioritize patients correctly (the one deemed the most urgent gets a faster appointment; called triage, in healthcare). The best way to get a quick appointment is to have used a telehealth service like First Derm, since a dermatologist has made a preliminary assessment that another dermatologist would respect.

The question is if AI could make informed triage decisions?

Artificial intelligence (AI) health chatbots medical data

Artificial intelligence (AI) health chatbots are considered by many to be the new health internet search. They are marketed as being more accurate and specific than a health internet search. Nevertheless in regards to a health internet search on skin diseases, a layman (the man on the street – non expert) compare the images that pop up from your search result to your skin concern. Instead of using an internet search to guess your skin concern, chatbots rely on correct text descriptions of your skin disease. They can ask up to 20 to 30 questions and from your answers refine the answer (it can take up to 5 minutes to complete a search).

How would the general public describe a skin disease? Health chatbots say that they have been trained on anonymous medical data from medical health records and can help hone in on the right diagnosis. The weakness here (if true), medical records have been written by medical staff such as nurses and doctors. The journal is written in medical terms, which in many cases is not how a layman would describe their medical condition.

Below are Tweet reactions on a recent study by Babylon Health (sponsored and written by Babylon doctors):

Babylon Health AI has been tested extensively by GP consultant @DrMurphy11. Here is one of his many tests

Click here to try out our FREE Skin Image Search

Artificial intelligence (AI) health chatbots vs. Skin Image Search challenge

Currently there are no health chatbots optimized for skin diseases that we know of. However a text skin search is possible in the respective chatbots. So we decided to test each chatbot and compared it to our Skin Image Search results, which has been trained on 33 different skin diseases that you can read more about here.

National Health Service (NHS) direct skin disease photos and text

We selected the images in this test from the UK National Health Service (NHS) Direct – it is considered to be a very good unbiased source of health information that is peer reviewed and sponsored by the UK Government. Keep in mind that the AI has been trained on skin disease photos taken with a smartphone, so we tried to find similar photos. On the left you can see the image and text accompanied by it underneath. On the right you can see the result with AI dermatology, Skin Image Search.

Atopic dermatitis

Source: http://www.nhsdirect.wales.nhs.uk/encyclopaedia/e/article/eczema(atopic)/

Atopic eczema (atopic dermatitis) is the most common form of eczema, a condition that causes the skin to become itchy, red, dry and cracked.

Contact dermatitis

Source: https://www.nhs.uk/conditions/contact-dermatitis

Contact dermatitis causes the skin to become red, blistered, dry and cracked. This reaction usually occurs within a few hours or days of exposure to an irritant or allergen.

N.B This image has been looked at by two senior First Derm dermatologists and do not agree with the NHS diagnosis contact dermatitis, but rather folliculitis.

Folliculitis

Source: https://www.nhs.uk/conditions/ingrown-hairs/

Ingrown hairs are hairs that have curled round and grown back into the skin. They produce raised red spots, which can sometimes become infected and turn into painful, pus-filled sores.

Herpes Simplex – cold sores

Source: https://www.nhs.uk/live-well/healthy-body/common-skin-conditions/

Cold sores are small blisters that develop on the lips or around the mouth, caused by the herpes simplex virus. They often start with a tingling, itching or burning sensation around your mouth. Small fluid-filled sores then appear, usually on the edges of your lower lip.

N.B The AI confused this as labia majora of the female genitalia, however it did get herpes correct. The correct answer should have been Herpes Simplex, which came up as alternative number 4.

Pityriasis veriscolor

Source: https://www.nhs.uk/conditions/pityriasis-versicolor/

Small patches of skin to become scaly and discolored.

Psoriasis

Source: https://www.nhs.uk/conditions/psoriasis/symptoms/

Psoriasis is a skin condition that causes red, flaky, crusty patches of skin covered with silvery scales

Basal Cell Carcinoma (BCC)

Source: https://www.whittington.nhs.uk/mini-apps/medicalinformation/MIPage.asp?id=14

A scab that bleeds occasionally and does not heal completely

Malignant Melanoma

Source: https://www.nhs.uk/Tools/Documents/mole_a_visual_guide.html

Appearance of a new mole or a change in an existing mole. This can occur anywhere on the body, but the most commonly affected areas are the back in men and the legs in women. Melanomas are uncommon in areas which are protected from sun exposure, such as the buttocks and the scalp. In most cases, melanomas have an irregular shape and are more than one colour. The mole may also be larger than normal and can sometimes be itchy or bleed. Look out for a mole which changes progressively in shape, size and/or colour.

N.B. This photo has been taken in a clinic setting with a high resolution digital camera and not a smartphone. 

Mole – Birthmark – Nevus

Source: http://www.nhsdirect.wales.nhs.uk/encyclopaedia/b/article/birthmarks/

Birthmarks are colored marks that are visible on the skin. They’re often present at birth or develop soon afterwards. There are several different types of birthmark and some of them are very common.

Genital Warts

Source: https://www.nhs.uk/conditions/genital-warts/

One or more painless growths or lumps around your vagina, penis or anus and can be Itching or bleeding.

Genital Herpes

Source: https://www.nhs.uk/conditions/genital-herpes/

Small blisters that burst to leave red, open sores around your genitals, anus, thighs or buttocks tingling, burning or itching around your genitals

Genital Molluscum Contagiosum

Source: https://www.nhs.uk/Tools/Documents/sore_penis_a_visual_guide.html

Spots or abnormal patches of skin on the penis could be a sign of molluscum contagiosum (MC). The lesions are usually 2-6mm across with a central dimple. They are usually firm, raised and painless and may appear in clusters. They may rupture to give out a yellowy-white substance.

Click here to try out our FREE Skin Image Search

Health Chatbots Text Tested

We used the same text and keywords from NHS Direct as seen in the skin diseases above. We have speeded the videos up to 2.5X, if a video is 1 minute, in real life it takes 2 min 30 seconds to see the final answer. We tested 12 skin diseases on 7 health chatbots. Only 3 videos appear per chatbot, but you can see the results in the table below.

ADA Health https://ada.com/

Founded 2011. Berlin, Germany. Total funding amount $69M. Ada Health offers a new approach to personalized health, supported by sophisticated artificial intelligence technology.

Babylon Health https://www.babylonhealth.com/

Founded 2013. London, England, United Kingdom. Total funding amount $17.3M. Putting an accessible and affordable health service into the hands of every person on earth.

Buoy Health https://www.buoyhealth.com/

Founded 2014. Boston, Massachusetts, USA. Total funding amount $9M. Buoy is an online symptom and cure checker that uses an intelligent algorithm backed by medical data to diagnose patients.

Infermedica https://symptomate.com/

Founded 2013. Wroclaw, Dolnoslaskie, Poland Total funding amount $5M Infermedica uses artificial intelligence for medical diagnosis and patient triage.

K Health https://www.khealth.ai/

Founded 2016. New York, USA. Total funding amount $19M. K Health is a health tech company empowering people with access to personalized health information based on “People Like Me”.

Sense.ly http://www.sensely.com/

Founded 2013. San Francisco, California, United States  Total funding amount $11.8M. Sensely is an avatar-based, empathy-driven clinical platform that helps clinicians and patients better monitor and manage their health.

Your.MD https://www.your.md/

Founded 2012. London, England, United Kingdom. Total funding amount $17.3M. Your.MD is an AI-based health information service platform and a marketplace of health service providers and products.

Click here to try out our FREE Skin Image Search

Over all Results

Some services are not available on all devices and appart from a somewhat tedious Q&A texting format. Our test show that ADA Health came out in front, Buoy Health number two, however we did not manage to test all skin diseases as there was a limit to how many times you could use their chatbot. Symptomate number three and Your.MD number four. Babylon Health, K Health and Sensely could not give an appropriate answer to any of our tests. To be fair, they do market their service for skin conditions. Detailed results below; the table is grouped as Inflammatory disease, skin cancer and STDs.

Symptom checkers

WebMD

Founded 1996. New York, USA. Acquired by Internet Brands 2017 for $2.8B
WebMD provides timely and credible health information and services to consumers and healthcare professionals.

Healthline Media

Founded 1999. San Francisco, USA. Total funding amount $161.5M
Healthline Media is the 2nd largest and fastest growing consumer health publisher in the world, reaching 85 million people a month.

Isabel Healthcare

Founded 1999. Great Lakes, Midwestern US
Isabel Healthcare provides a diagnosis decision support tool that doctors rely on for their own second opinion on patient care.

NHS Wales Symptom Checker

The NHS has a symptom checker on potential skin cancer, that you can try out here

http://www.nhsdirect.wales.nhs.uk/SelfAssessments/symptomcheckers/?ScName=Mole2&SCTId=166

References

[1] Julia Schofield, Douglas Grindlay, Hywel Williams. SKIN CONDITIONS IN THE UK: a Health Care Needs Assessment. Centre of Evidence Based Dermatology, University of Nottingham. 2009.
[2] Lim HW, et al. The burden of skin disease in the United States. J Am Acad Dermatol. 2017 May;76(5):958-972.
[3] Ramsay, D. L., and A. B. Fox. “The Ability of Primary Care Physicians to Recognize the Common Dermatoses.” Archives of Dermatology 117.10 (1981): 620-22.
[4] From different public sources