sungam

I agree - personally I do not find the descriptive terms and language used to describe features on dermoscopy helpful to me in making a diagnosis, however I think they can be helpful when teaching other humans how to learn the technique. They may not provide additional benefit for machine learning classifiers which can deduce these principles directly from the images themselves.

This is the plan!

I think the reason I am surprised is that they are exceptionally good at interpreting images generally and there are tens of thousands of labeled skin cancer images on the public internet so I had assumed the general image interpretation abilities would transfer

Would be difficult to fit in a browser as state of the art requires multiple vision transformers. Also it would not be sensible to provide access to models without human oversight as at this point in time all models have the potential to miss skin cancer.

This does make sense, however in view of the recent maths results and since there are tens of thousands of training images for skin cancer available on the public internet I was expecting the LLMs to perform a lot better than they did

The CNN architecture used here is definitely outdated. I used it as a baseline to compare to the LLMs and because the model is small enough to run using WebGPU in a phone browser for demonstration. Modern approaches use ensembles of vision transformers and have proven accurate in clinical trials

I have been incredibly impressed by LLMs for coding and data analysis and was expecting the LLMs to perform better so I was surprised by the results especially since this is an old model and nowhere near state of the art for skin cancer diagnosis.

This is fantastic and I would do something similar if I had the same resources.

Whole genome sequencing and single cell sequencing are actually surprisingly inexpensive (compared to the cost of almost any cancer treatment) and there is a good argument that we should do this by default for any type of cancer for which there are not effective treatments.

I would encourage all patients to learn as much as possible about their own diagnosis. There may be clinical trials available in another centre that your clinician is not aware of or tumour-agnostic trials that target specific mutations that are present in a range of different cancers.

There is a good argument for allowing patients to try experimental treatments once standard treatments are exhausted. The provider liability issues could easily be solved by legislation. A bigger issue is that there will always be people who want to exploit vulnerable cancer patients by charging exorbitant amounts for plausible-sounding treatments which have no evidence base.

As an individual trying multiple experimental treatments at once is the logical approach if there is no other option. However it will not be possible to know which of these is effective and which have caused side effects so to develop more effective treatments we do need structured and carefully controlled clinical trials. Unfortunately there is a huge regulatory burden for any kind of clinical trial at present and this should be massively streamlined for cancer patients where even if the treatments cause harm the alternative is death.

Performing the procedure requires a high level of skill in interpreting scans (angiograms) in real time.

In fairness, while I’m sure it didn’t work out for all of them, one of these students became one of the youngest self made billionaires in history

This is true, however I am sure that AI-based image analysis will also replace some of the work of dermatologists eventually so nobody is safe...

Not at the present time. I think if the project takes off I will move it to static hosting anyway to reduce the attack surface.