Why AI in radiology failed
It was around 2018 when pressing news came that AI would soon be able to replace radiologists (mostly led by then publicized IBM Watson. But not only.).??
Doctors themselves dropped in mass (US data) taking Radiology as a specialization.
Well, in this case, AI failed. And we may have an actual shortage of radiologists.
Yet today AI scores on accurate diagnosis is about 65% while 85% average for a radiologist (see the paper from BMJ https://www.bmj.com/content/379/bmj-2022-072826)
What happened??
First, I wrote it implicitly above, the accuracy is (rightfully) calculated on diagnosis, not just images. A radiologist today provides diagnosis not just based on the radiography image alone but based on data and the general condition of the patient which is not really considered by the AI model.?
Second, the radiographs and its metadata are taken from different manufacturers and they are not standardised and sometimes labels are not strictly consistent. The AI model has been trained with heterogeneous data and not surprisingly the accuracy can be improved.?
领英推荐
The case moving forwards.
Overtime,? the second point above will be solved. Maybe 5 years? Maybe 10.It would depend how quickly data can be standardized and become ' AI friendly': the ease on which radiographies and their metadata can be ingested into AI training. Manufacturers may or not have the right incentives to do so.?
Still to be tackled how to make full diagnosis relating different patient data together with the radiography. Combining these data may return a variability of scenarios which may not have evenly distributed data so the model cannot actually replace the radiologist. In other words, until we have standardized data of patients with good statistics for different cases (to cope with variability of scenarios) it may be difficult for a model to perform as a human.?
My guess.
In the next 5 years, if there is an economic incentive, it may be possible to make a specialized AI model just for certain conditions but not in general (like a model just for cancers or broken bones). Essentially the radiologist just will change over time, not disappear.
In other words, an AI radiologist is actually closer to general ‘low’ AI level than just narrow AI. I stress ‘low’.? A human today can ingest/generalize across non-standardized heterogeneous data,? in a way we need yet to develop/invent in AI. If that happens, my guess would be wrong and AI radiologists would be around soon.
#ai #artificialintelligence #business #innovation #health