Artificial Intelligence skin lesion diagnostic apps-good or bad?

A position statement appeared in a recent Journal of the European Academy of Dermatology and Venereology (JEADV) was recently published. It was a thoughtful and thorough piece of work by the EADV Task Force on Artificial Intelligence (AI), led by various luminaries of the skin cancer diagnostics world including Professor Harald Kittler, whom regular readers here will know is held in admiration for his brilliant mind, body of work and devotion to skin cancer diagnostics research. It is an open access article which can be freely read and downloaded.

Read it yourself if you are interested, but in summary the statement says that Artificial Intelligence in dermatology, including Smartphone apps, is already here and has great potential for good, but could also do some harm if we’re not careful, so we ought to keep an eye on it to make sure it is done right and not used to frighten and exploit people.

I thought of this as I came across the following case from my image archive, which I am currently editing and setting in order since I have now retired from clinical practice and wish to pass on when I pass on. This case was of a young-ish person with a long standing mole on a limb. They took a picture of it and fed it into a mobile phone application, which said it was worrying and told them to seek advice. They went to their GP, who unsurprisingly made an urgent skin cancer referral, although there was no history of recent change or other pointers to a skin cancer.

This is a mole of slightly above average size on a white adult, larger than the other 5 or 6 moles in view, but not dramatically so. Importantly, it had been there for years and there was no recent change. So far, I am not very anxious, but let’s take a closer look.

It’s not a great photo, but doesn’t raise any red flags. Let’s get the dermoscope out.

This is really a very bland mole. It is essentially featureless, with fairly good symmetry (the slightly darker area at the top too trivial to count), a little reticular network around the periphery from 10 to 3 o’clock, fairly even border and no positive melanoma clues. My judgment was that this was a bland naevus with no worrying features and could be safely observed.

However, by this time the patient was frightened. They had been initially scared by the AI Smartphone app, then by their GP’s failure to reassure (not surprising since British GPs get about half an hour of skin lesion recognition training in a lifetime, they all live in fear of being sued or complained against, and they have nothing to lose by referring all such cases to hospital) and of course they had been summoned urgently to the SKIN CANCER clinic with a letter that had the word CANCER all over it. They were not prepared to accept my reassurance.

The lesion was duly cut out and examined by two of our expert histopathologists, who duly confirmed my informed and experienced judgment that this was an entirely harmless naevus, not even ‘mildly dysplastic.’

SO, a lot of fuss and trouble, and a permanent scar, because of a for-profit Smartphone app which had never been signed off by any medical authority and will UNDOUBTEDLY have been set at an extremely high sensitivity/low specificity diagnostic level in order to make sure the app’s makers will never, ever be sued for a missed melanoma. ‘Better safe than sorry’ as the saying goes? They won’t be the ones dealing with the needless anxiety and extra work generated by their profitable product.

OK, that’s one story. But it’s not the only one I have come across, and Alan Halpern, a leading New York skin cancer professor, said at the EADV in Milan in September 2023 that they were already seeing big problems in America and Scandinavia due to overdiagnosis by these AI devices. The professor said ‘These Smartphone skin lesion diagnostics apps are going to send us a huge amount of extra work.’ In a health care system where more work means more item of service fees and increased profits, they will not be weeping much over this. But it’s a different matter in a cash-limited service like the British NHS where we have rationing by queue. If more people join the queue, it gets longer.

I agree with the EADV Task Force that there are benefits to be had from AI dermatology Smartphone apps. But as with everything else that clever (but not always wise) humans come up with, there will be a down side as well. Let’s be careful out there. As I write in January 2024, Britain is reeling from the Post Office false accounting scandal, which was caused by a bad piece of software and by managers promoted above their level of competence who refused to entertain the possibility that the software could be wrong. It was wrong, badly, and many lives were ruined.