Data driven diagnosis: ODDSPOT

Diagnosing suspicious spots on your skin for evidence of Actinic Keratosis and Basal Cell Carcinoma

From the scientific literature we know that some decisions can be adequately supported by or even made by a computer model. That might take some getting used to (“how can my phone be just as smart as a human expert”), but in the Data Science world this is actually common knowledge since at least 10 years. It is also something one can expect, given that computers nowadays are really good at well-defined tasks such as playing chess and checkers, and recognizing faces.

To showcase this, we implemented a decision model that is able to quite accurately diagnose two kinds of suspicious spots on the skin: Actinic Keratosis (AK) and Basal Cell Carcinoma (BCC). From all suspicious spots on the skin, AK and BCC are among the most commonly occurring.

Oddspot FAQ

How good is the app?
The short answer is: quite good, namely better than your general physician will be in diagnosing AK or BCC, about as good as a trained nurse in the dermatology ward, and slightly less good than a trained dermatologist.
Actually, this is a really hard question to answer generally. We know for instance that general physicians are not very good at correctly diagnosing such spots specifically, but this is not really their task. The general physician has to know when it would be wise to involve a specialist, and they are generally quite good at that. You should interpret the app’s output in the same way: as a helpful hint for you to decide whether or not to consult a general physician or dermatologist, and it is pretty good at that.

Does the app make mistakes?
Undoubtedly. As does your general physician and dermatologist. Luckily, this does not happen often (see the previous question), but it can happen. So when in doubt, do consult your physician or a dermatologist, so that they can decide whether or not further testing (for instance a biopsy) is necessary.

So I don’t need my physician anymore for suspicious spots?
You still might. This is a decision-aid, not a physician in a box. The app only checks against AK and BCC, and there are obviously other potential diagnoses, such as eczema (which can take different shapes and sizes) or melanoma (which typically looks like an irregular mole). When in doubt, do consult your physician. When you work with or are related to a hospital or dermatology ward, and are interested in further developing the app so that it can diagnose more together, please contact us at c.c.p.snijders@tue.nl.

How was the app made?
We first browsed the literature and asked dermatologists which factors are suggestive of Actinic Keratosis or Basal Cell Carcinoma. We then measured these factors among a couple hundred persons, and figured out which of these factors best predict actual AK and BCC. Sounds simple, and it does work.

Who created the app?
The development of het underlying model is the result of cooperation between two scientists with expertise in decision-making (Chris Snijders, TU/e and JADS; Ad Kleingeld TU/e) and dermatologists Gertruud Krekels and Simone van der Geer (MOHSA). Several students have participated in the data collection, first analyses, and pilot versions of the app. The current version has been implemented at Jheronimus Academy of Data Science (JADS).

Are my data anonymous?
Besides the data you enter yourself, nothing is being saved in the database (no ip-address, no cookies, nothing). Usage of the app is completely anonymous. The questions contain nothing that can be traced back to you. We value your privacy.

Watch the video about this case (in Dutch)