Futurology

This is what AI should be used for. Not the generative crap ChatGPT peddles.

AI is perfect for applications looking at tons of different variables for specific patterns and are capable of being trained on new data cheaper than training every doctor in the country.

A doctor's first and primary goal is keeping a patient alive. Second is to normalize quality of life. Third is to minimize suffering when possible.

There is a HUGE and artificial shortage of doctors and healthcare providers in this country, and largely the world. They honestly don't have enough time to review every patient record, symptoms, and make a diagnosis and treatment plan, THEN do their continuing education and licensing requirements, AND do any research if they are mandated to do so by their employer, AND if they are at a teaching hospital - teach.

These AI tools can look at an entire medical record, symptoms, laboratory results, and pathology images and make a very accurate diagnosis that is always run by a physician before making a determination. AI doesn't forget what it's learned either.

I really wish “AI” would die; machine vision and convolutional neural networks used in this application don’t have much to do with the large language models most people think of with the modern incarnation of the term ai

Here's the paper: https://www.sciencedirect.com/science/article/pii/S2666990025000059?via=ihub

The confusion matrix and ROC curve are in section 5.2.

The image processing pipeline includes techniques from the 00s (in preprocessing such as otsu and watershed), to quite recent (gated MLP "transformers light").

I am able to identify 100% of cancer: just say "It is cancer" to each picture.

~~The article does not mention any other metrics than detection rate. What about recall etc.? Without it, this news is basically worthless.~~

I stand corrected, see the comments below. While the article still lacks important context, accuracy is well defined for this topic.

one of the particularly good uses for AI! in fact it's so good and cheap that it'd actually be hard to turn a lot of profit on! which... hm....

If I state "every living creature that ever existed or will ever exist had, has, or will have cancer" I just diagnosed all the cancer in existence ... including cancer thousands of years from now. That is a 100% diagnosis rate.

But what would be the error rate?

From the article: Of course, it's not a tool designed to replace medical professionals but to be used in collaboration with cancer specialists to accurately spot the disease and then monitor how successful treatment has been. What's more, this kind of model is a much more rapid, accessible and affordable way to diagnose cancers.

This is the key difference and how AI should be used. It doesn't replace the human but effectively aids them in their research. The whole "outperfoming doctors" pitch needs to change to "Reducing critical misses for doctors." Otherwise it gets roped into the ChatGPT-like AI's which are absolutely garbage for decision making.