After a short intermezzo last summer, the European Congress of Radiology returned to its traditional spot early March in Vienna, Austria. Attendance reached its highest level since before the pandemic, with 17,262 participants from 122 countries, a 14 percent increase compared to ECR 2022. In the technical exhibition, which featured over 250 medical imaging companies, two topics clearly dominated the show: artificial intelligence (AI) and sustainability.
Companies showed that they are not only concerned with providing their customers with energy-saving systems, but also placing a lot of emphasis on reducing energy consumption along the entire value chain, saving resources, and generating as little waste as possible. I remember reading somewhere: “Did you know that healthcare is responsible for 4.4 % of global CO2 emissions? That is more than the entire aviation or shipping industry consumes!”
When it comes to AI, where to start? There are so many and varied individual applications. It is best if we just try to classify them. On the one hand, we have workflow AI, which is about speeding up administrative processes and boosting workflow efficiency. Workflow AI starts with making appointments, continues with registration and the filling out of examination-specific forms, and ends with the automated creation of reports. In between, so-called workflow orchestrators ensure
that decisive information is always at the right place at the right time. Each radiologist gets to see the images that he or she can best assess. At the same time, the system makes sure that the radiologists’ work lists don’t become too long, and that patients with clear findings are given priority. All in all, the aim is to individualize, standardize and automate workflows.
Pixel AI is the area that is currently most widely associated with AI: clinical decision support, i.e. the automated recognition of findings. The question arises again and again whether this form of AI will soon replace radiologists. But those who ask this question have not yet fully understood the task for AI. In clinical decision support, AI is used to free radiologists from repetitive, error-prone tasks. Normal findings, which make up the majority of all radiological examinations, can reliably be recognized by AI. This leaves radiologists more time to deal with difficult cases, possibly also together in conferences. Unfortunately, the number of examinations is constantly increasing due to demographic developments. The global population is aging and growing at the same time. By 2050, there will be more people older than 70 living on earth than the world population in 1950.
Furthermore, AI algorithms can increasingly be found directly on the devices, as showed at ECR 2023. Cameras evaluate the body contours to position the patient correctly inside the CT scanner by just pushing a button, and at the same time suggest the right examination protocol. In MRI, sequences are accelerated and scan times are significantly shortened. Ultrasound systems automatically recognize findings and independently optimize scan parameters. There are hardly any limits to the technology.
The possibilities for mastering the challenges of modern radiology with the help of AI are manifold. But there are still one or two downers: for example, the validation of the different algorithms. This currently takes place exclusively on the basis of clinical studies. It would be desirable to have a technical check similar to the one that is commonly done with X-ray and CT systems: put the test specimen on the table, take a picture or scan, have the image assessed by the algorithm, and that’s it.
But this may take a while. Committees across Europe are still positioning themselves and must first agree on systems’ validation. But there is hardly a better time to start with AI than now – even if everything isn’t perfect yet. The path from single slice step-and-shoot to photon counting CT also took several decades.
I’m already looking forward to ECR 2024, with its exciting theme Next Generation Radiology.