[...]

[fa21212]

[...]

---

Members don't see this ad.

 

[LoGo]

For derm, if anything it will help expedite diagnosis (maybe fewer referrals) and minimize the need for biopsy. The reason is that you still need a dermatologist to either (a) say what the lesion is if it's NOT what the AI is looking for, and (b) you need someone to actually act on the diagnosis be it surgery, medicine, laser etc. Further, there are too many rare Dx in derm that it would be nearly impossible to gather enough training data to develop an AI system that is capable of outperforming a dermatologist at everything.

 

[TMP-SMX]

That's great color, thank you. What percentage of billings from derm comes from diagnostics vs. procedures? Do you have a sense at all?

For dermatology surgeons (mohs) obviously it's skewed. For a general dermatologist that doesn't do as much surgery (most do quite a bit but some are specifically medical dermatologists with limited procedures) there might be a higher amount of high level E&M codes over destruction/biopsy. Most have a good mix. There's still a lot of psoriasis, acne, and eczema out there that require office visits. On the other hand there's plenty of actinic keratoses, warts (and inflamed seborrheic keratoses), and skin cancer out there as well.

My opinion is similar though. Someone still has to treat the lesion and chose which one is relevant. It would certainly save us for important consults aside from reassuring the benign nature of seborrheic keratoses. The major downside of tele dermatology and AI at this point is it is not rapid enough and high resolution enough to make a good clinical diagnosis. Sure, a medical photographer could take a lot of photos but it's not always known if the area photographed is the most representative etc.

 

[werd]

i'd be more impressed with the study result if it compared US-trained and practicing dermatologists to the AI in melanoma detection. I'm really not sure what the training and standards are like for dermatologists in many countries of the world. also, correctly identifying a lesion of question is different from evaluating hundreds of lesions on a patient in a short period of time, as is done during a "skin check." there's also the point above that there are many very rare and bad cutaneous malignancies, so much so that it would be difficult to train any AI to identify them readily (Merkel cell carcinoma, pleomorphic sarcoma, angiosarcoma, aggressive digital papillary adenocarcinoma, AFX, microcystic adnexal carcinoma, DFSP, Paget's disease, leiomyosarcoma, sebaceous carcinoma, etc..). I think, in general, this sort of technology is a positive development. Technology may replace the expertise of dermatologists for lesion identification in some fashion in the future; on the other hand, we can also learn from the AI by finding out what it is "picking up on" to identify melanomas that even dermatologists miss.

 

[Redpancreas]

So I think the potential disruption of diagnostic radiology from AI has been discussed at length but I haven't seen as much on dermatology. This article speaks to a little bit about computer augmenting / replacing the diagnostic function of dermatologists. In addition, I'm sure you saw the news recently that AI outperformed experienced dermatologist in recognizing melanomas. How do you guys think about potential disruption in these fields and other fields of medicine from deep learning?

A.I. Versus M.D.

AI better than dermatologists at detecting skin cancer, study finds

Derm’s not going anywhere /endthread

 

[FindMeOnTheLinks]

So the computer has a high sensitivity and specificity in identifying melanoma and might be a bit better than humans. Great. But the computer needs parameters - somebody still has to tell it what it's looking for.

A CT scan has 94% sensitivity and 95% specificity for appendicitis (similar to the AI in diagnosing melanoma). It is undoubtedly better than human clinical judgement in diagnosing appendicitis. But a CT scan only works if you order it.

The point being, the AI is just another diagnostic tool to augment clinical judgement, like any other diagnostic test. Yeah maybe it's a pretty good test, but it doesn't work in the absence of a human who can tell it what it should be looking for.

 

[ndafife]

Someone has to cut out the skin lesion even if AI becomes the standard of diagnosis. And surgery pays better.