Hello everyone and welcome to ZD Tech, ZDNet’s daily editorial podcast. My name is Clarisse Treilles and today I explain to you why artificial intelligence has not eradicated radiologists.
In 2016, the British Geoffrey Hinton, one of the pundits of artificial intelligence (AI), predicted that the progress of AI in medicine would quickly make the profession of radiologist disappear. Not without a touch of provocation, he declared, and I quote: “In five years, deep learning will be so powerful that it will no longer be useful to train radiologists. »
But five years later, that hasn’t happened. Quite the contrary. The World Economic Forum even points out in a study published in the midst of the Covid-19 pandemic that the number of radiologists has always been experiencing double-digit growth for decades.
It is therefore that humans remain indispensable for diagnosing fractures and cocaine balloons in the intestines of traffickers. In short, AI has not made the profession of radiologist and customs officer disappear, far from it. But AI has still changed the practice quite a bit. Listen, I’ll tell you how.
AI is not infallible
Already, the AI is not error-free. We cannot rely 100% on algorithms to interpret X-rays and make the right diagnoses, especially when the pathologies are complex or rare.
Machine learning techniques indeed require hundreds of thousands or even millions of images of different cases to improve the detection algorithms. And that takes time.
But artificial intelligence is definitely helping to change the way radiologists work. Judge instead. In a hospital, for example, the algos make it possible to make the first level sorting between the most benign cases and those which deserve a human eye.
This is exactly the playground of the French start-up Milvue, which ZDNet already talked about last year. Its RN takes care of routine x-rays, allowing radiologists to focus on images that show more complex pathologies.
The augmented radiologist
So how does it actually work? Milvue has developed an algorithm that retrieves X-ray images from the emergency room, interprets them, and makes them available either to the emergency physician or to the radiologist, classifying them according to three criteria: normal X-rays, questionable X-rays or X-rays considered pathological.
And this AI has also been tested in much more specialized specialties. In the detection of breast cancer, for example, research is promising. The journal Radiology says AI is likely to be of great value in interpreting screening mammograms.
The fear that AI will kill the profession of radiologist has therefore largely dissipated as the capabilities of artificial intelligence are better understood. And above all, what we must remember from Geoffrey Hinton’s prediction is that we should never blindly trust technological progress.